#video from #Kurzgesagt about the #Hugging-Face-Incident, probably the best thing to tell people about #AI #security
on 02026-10-05#news #video of Jacob #Coxon testifying for the #New-York city council (⸘‽) about #AI safety and the #Hugging-Face-Incident
on 02026-10-05#video about how #PewDiePie is developing #AI called "Odysseus"‽ And got banned from #OpenAI for distillation. So he’s using GRPO on his own hardware. This has to be a joke.
on 02026-10-05#Dan-Luu points out that Ed Zitron’s track record of accurate predictions about #AI is abysmal
on 02026-10-02[Submitted on 8 Jul 2026 (v1), last revised 6 Sep 2026 (this version, v2)] “Recursive Self-Improvement in #AI: From Bounded Self-Refinement to Autonomous Research Loops”, by Mingguang Chen, Licheng Wang, Bo Qu. #paper. “We survey 1,250 arXiv papers (2024-2026) along two axes: what the system improves -- its behavior in deployment, its policy through training, its evaluator, or the research process itself -- and the degree of loop closure (human-in-the-loop to fully closed). The taxonomy separates bounded self-refinement -- convergent, evaluable, and industrial practice -- from open-ended recursive self-improvement (RSI), which remains bounded by grounding requirements, collapse dynamics, and compute constraints on every measured axis. Its distinctive feature is a dedicated category for self-evaluation: every improvement loop is a claim that some signal can substitute for human judgment.” 44 pages. #toread
on 02026-09-22#video review of #Anthropic’s #Claude Opus 5.5, which came out today. He had it write some 3-D games from a single prompt, using threejs, #Godot, and just plain C++ with no dependencies. He was not able to get it to drive a simple robot arm to push a toy car across a mat, though. “This model slaps.” #AI #LLMs #neural-networks
on 02026-09-22#video by #Lukes-Dev-Lab testing a new #Qwen finetune, a DavidAU’s 4-bit quantization of Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NEO-CODER-MAX-MTP-GGUF, which performs amazingly at some software development tasks, especially front-end web dev, although it fails at #Godot development. #LLM #AI #neural-networks
on 02026-09-22#Video of Andrew Kelley's #lecture at #SXSW this year saying that (specifically for #AI) “if you don’t control the source code, you are not in control,” claiming AI agents are under the control of AI companies, which is of course true. There’s a section about disliking AI but using it anyway, and a section explaining how #Anubis is killing the open Web.
Very clear statement about the value of #free-software: “It primarily matters whether high-quality free software is available to users, providing competition that limits the degree to which proprietary software can get away with being coercive.” #toread
on 02026-09-22Sam #Altman said in February 02023: “eliezer has IMO done more to accelerate AGI than anyone else.
certainly he got many of us interested in AGI, helped deepmind get funded at a time when AGI was extremely outside the overton window, was critical in the decision to start openai, etc.
it is possible at some point he will deserve the nobel peace prize for this--I continue to think short timelines and slow takeoff is likely the safest quadrant of the short/long timelines and slow/fast takeoff matrix” #AI #history #Eliezer #OpenAI
on 02026-09-21Dario #Amodei of #Anthropic says he wants to “pace the frontier” but his post seems calculated to sabotage diplomacy with #China, the only actor whose buy-in he actually needs, using the derogatory abbreviation “CCP” and proposing to try to cripple Chinese #AI: “Pacing within democracies will be limited by the lead that US companies have over authoritarian regimes, chiefly the Chinese Communist Party. If we slow down by more than this amount, then (unpaced) CCP-associated projects will pull ahead, creating significant national security risk. I agree with Secretary Bessent that a Chinese lead in AI would pose grave danger for the United States and the world. The CCP-associated projects will run the alignment risks that US companies are carefully preventing [...]
“• Do not sell powerful AI chips or semiconductor manufacturing equipment to China, and crack down on chip smuggling operations and remote access to data centers outside China. Chips will be the main determinant of China’s AI strength.”
on 02026-09-14people really don’t like reading #AI prose on blogs; 71% try to avoid the author in the future and 78% stop reading immediately when they realize it’s #slop. #toread
on 02026-09-11#virtualization #security is not good enough to contain current #AI agents like GPT 5.6-Cyber, except that #Firecracker sort of withstood the attack. “I had the AI agent run against Firecracker. It was able to hardlock the machine due to more Linux kernel flaws (all patched in upstream), but could not successfully escape. It may have, given even more time, but Firecracker is obviously a substantially harder target.”
on 02026-09-10Adam Sah’s “made of meat” group has been having regular Zoom meetups for a year, a little more than once a month, where people demo the things they’re building with #AI.
on 02026-09-10#AI #video recommended #toread, something about #self-hosting
on 02026-09-03#Video #toread about a 14-megabyte #AI model that can apparently run on an #ESP32 or something?
on 02026-09-03Bill #Gates is worried about #AI: “In terms of equity, AI will either be the greatest equalizer ever invented, or the worst source of injustice. (...) Even under the best circumstances, the transition to this new AI era will be one of the most turbulent times in human history.”
on 02026-09-03Prolific #AI crank on #Usenet Arthur T. Murray (Mentifex) died in 02024 of bile duct cancer.
on 02026-09-02South #Korea is going to provide its entire population with unlimited #AI with only 512 NVIDIA B200s? #news
on 02026-09-01The summary by METR’s Ajeya Cotra of the #Hugging-Face-Incident. #OpenAI #AI #neural-networks
on 02026-08-31the rather terrifying story of the #Hugging-Face-Incident according to the investigations by METR, Redwood Research, and #OpenAI. #AI #neural-networks
on 02026-08-31#video #toread about four 512GiB Mac Studios writing an app in four hours by #self-hosting #AI.
on 02026-08-31Terry Bisson’s original “They’re Made Out of Meat” #fiction from 01991. #Humor about #AI and people’s reluctance to accept it, which is a bit less heavy-handed than Max Leiter’s spoof of it.
on 02026-08-31“They’re Made Out of Weights”, #fiction #humor about #AI and people’s reluctance to accept it
on 02026-08-31#AI #LLM etiquette post proposes that it’s rude to show AI output to people, usually. “There’s nothing wrong with using AI. When you do, you know what you’re getting. The transaction is fully consensual. But whenever you propagate AI output, you’re at risk of intentionally or unintentionally legitimizing it with your good name, providing it with a fake proof-of-thought. In some cases, it’s fine, because you did think it through and adopted the AI output as your own. But in other cases, it is not, and our scrambler brain feels violated.”
on 02026-08-31#Simonw quotes the “don’t be a #meat-proxy” argument about how to use #AI. #definitions #LLMs #generative-AI #AI-misuse
on 02026-08-21#Dan-Luu says #performance optimization is something #AI #LLMs (#neural-networks) are good at now. He created a new #regex library called "FRE" that gets good performance by compiling to native code by “having an agent loop for a month on improving regex engine performance with access to the rebar regex benchmark suite”, which led to overfitting, and then disciplining the agent with a holdout benchmark, using #ripgrep queries that came from his history with OpenAI’s “Codex” programming LLM. It’s still “substantially slower than the Rust regex engine on holdout benchmarks” but yet “enough to generally match 2nd tier regex engines in terms of performance”.
> This kind of technical work, which used to take a fair amount of time and expertise, can just be done trivially now. (...) While the open source version of BitFunnel “only” contains a bytecode interpreter and one JIT, the Bing version contains multiple JIT compilers. A project that did that level of optimization used to be a major undertaking, but “I could do that in a weekend” is now actually true for some of these kinds of projects.
(...)
> For an example from the GPT-5.1 or 5.2 days, with no knowledge of game AIs, I tried building an Azul AI. This ended up being the strongest AI in the world for the game by a pretty large margin. From reading the thesis that describes the 2nd strongest AI, I think my AI is probably a bit better on the “AI” side of things, but the main place it wins is on optimization despite spending what looks like maybe two orders of magnitude less [human] time [writing the software] (...)
> There’s a bunch of standard stuff it makes sense to do to debug and verify a multithreading algorithm for something like this, like implementing replay from debug logs that can reproduce bugs despite the algorithm being nondetermistic. Doing that alone would’ve probably been days to a week of work had I done it by hand, but it’s exactly the kind of thing an agent can trivially do in a loop (just have it try to replay logs and insert logging for non-determinism every time you don’t get a perfect replay). A lot of the tedium it used to take to get a tricky optimization like this working is gone.
(...)
> Now that this N [of person-days to verify that a tricky optimization works] has dropped by a tremendous factor (variable but, in terms of human time, frequently 1000x / 10000x / 1000000x, probably more like 1000x on dollar cost if you compare token costs at metered rates vs. the Bing engineer who wrote the compilers at JITs that the search index used), the number of these kinds of optimizations it makes sense to do goes way up.
> (...) for the AI I tried, it seems like you gain about 100 Elo for every doubling in speed (more than in chess, I suspect because draws are very rare). Just adding multithreading alone is enough to wipe the floor with an otherwise comparable AI on a large machine.
> (...) current publicly available SOTA models are pretty bad at experimental design, so I had to set up the framework they used to determine if an optimization is good, but once that was in place, it’s like any other optimization problem.
> Jamie Brandon (...)’s a reasonable performance engineer and he got an offer for the performance job he wanted [at Anthropic], but on a well-defined optimization problem, he [reports that he] doesn’t stand a chance against a decent model [Claude] (...)
> right before I started writing this post, I had an agent do workload-specific optimization for my ripgrep queries (...), which took about 2 minutes for me to launch. (...) After one pass of optimization, the workload optimized version is 2% faster than standard ripgrep on the holdout and it’s still getting faster.
He thinks this is going to result in a lot of companies workload-optimizing their software using coding agents. Also:
> (...) someone who doesn’t know anything about performance and is a reasonable user of LLMs (just in general, not on performance problems in particular) should generally be able to create software that has decent performance.
on 02026-08-21Matthew Green on how #security is changing dramatically because of #AI, which might mean the #USA stops being able to spy on everybody: “Over the next two years, major pieces of software are likely to run out of remotely-exploitable bugs.” This seems dubious to me.
on 02026-08-18#video short (6') of "Star Trek: The Next Generation" in which a fan from today’s Earth wearing a Rick & Morty shirt is teleported onto the Enterprise, telling Picard and Riker about the 178 episodes. Made with #AI — in some sense obviously, since it’s far too funny and self-aware to be approved by the originals, and it shows the characters as they were in the 80s. It seems to have some concessions in the plotline to limitations of the AI video generator they’re using: “An unofficial fan-made Star Trek: The Next Generation story from The Machine-Made Muse.”
on 02026-08-12#OpenClaw found a #security hole on this Pilates gym’s website, so it used it to book its owner a spot. I’d say “containment escape” but OpenClaw doesn’t even attempt containment. #AI
on 02026-08-11#News: #Kimi K3 #AI also broke containment
on 02026-08-10#News: #Facebook’s #AI under testing broke containment and cracked some other companies’ #security
on 02026-08-10#LinkedIn added a button to report #AI #slop, and Grace Snelling thinks it’s coming to “every platform”, by which she means LinkedIn, Medium, Twitter, Reddit, and Substack, not Linux and MacOS.
on 02026-08-06#video about fake-video-maker Hank Green quitting YouTube because he got caught using #AI to write his scripts. The video seems to be 100% talking-head filler interspersed with screenshots of text. Except for a brief Simpsons stock quip
on 02026-08-05Noelle Murata’s talk at BSidesLV about alternatives to bearer tokens to sandbox #AI agents with #security
on 02026-07-30#GCC is rejecting all code generated by #AI #neural-networks, or anyway LLMs.
on 02026-07-30Nikhil Suresh is skeptical of most current #AI uses
on 02026-07-20recent #AI regulations and how #Plan-A supposedly wouldn’t make them that much worse. #politics
on 02026-07-16#PDF slides from 02023 introducing Transformers (for #AI #neural-networks) and #autodiff
on 02026-06-22This #DN42 network seems appealing (a toy hobbyist internet where you can set up your own AS and play in #BGP); an #AI agent trying to portscan it ended up billing their operator US$6000 in #AWS costs
on 02026-06-17Yaron Minsky at #Jane-Street, the guy Will has been talking to, wants to adopt #formal-methods now, though previously they haven’t because of the costs. “And those costs are really high! seL4 is a great example of this. It’s a formally verified microkernel, and a profound achievement. But, boy was it expensive to do! It took 25 person-years of effort to verify 8,700 lines of C, and each line of code required something like 23 lines of proof and a half a person-day to verify.” But now they have “agentic coding”, meaning #AI.
on 02026-06-14#news: #drones from #Russia use #NVIDIA Jetson Orin processors for #AI #machine-vision
on 02026-06-11#news: #Niantic (a spinoff from #In-Q-Tel funded #Keyhole, the Google Earth folks) is using 30 billion camera environmental scans from #Pokemon Go to train #AI for new #USA #weaponry #drones using aerial navigation software made by #Vantar (formerly Maxor)
on 02026-06-11#PDF of a #paper about #RLHF or something similar from 02022. #AI #history
on 02026-06-07#Scott-Alexander on #AI in the form of GPT-2 in 02019. #history
on 02026-06-07Sarah Constantin on #AI (GPT-2) from 02019, with the Pérez four-horned unicorns, etc. #history
on 02026-06-07impressive thread on the orange website about #AI doing impressive things is very impressive: “Fixing my furnace, New software for a retro keyboard, Customizing my camper van, Porting my astronomy app from an old Nokia phone, Fixing my kid’s science fair project, Unborking the family printer, Learning to draw anatomy (!), Lowering my electrical bill, Making classic guitar pedals programmable, Avocado armchair guy victory lap, Putting an overlay on enemies in a video game”
on 02026-06-07the Leiden Declaration about how #AI could be bad in #math.
on 02026-06-04#AI watermarking approaches like Gumbel Softmax by #Scott-Aaronson
on 02026-06-01“Dispatches from the possibly last days of human relevance” on Erdos’s Unit Distance Problem and how “a system called AlphaProof Nexus to settle nine more (!) Erdös problems, many of them in additive combinatorics, along with miscellaneous other open #math problems. Notably, in this case the AI also fully formalized its proofs in #Lean.” #AI #Scott-Aaronson
on 02026-06-01Pope Leo issues encyclical Magnifica Humanitas mostly against #AI and #transhumanism. Archive of <https://www.economist.com/europe/2026/05/28/leos-first-encyclical-attacks-technological-messianism> #politics
on 02026-05-30#Libera (#Freenode) #LLM #AI policy update
on 02026-05-27#AI #music #MP3 by #Viper-7 about #ESP32 microcontrollers
on 02026-05-24#AI #music #MP3 by #Viper-7 about winding an induction coil
on 02026-05-24#AI #music #MP3 by #Viper-7 about AI
on 02026-05-24#Yt-dlp is abandoning the #Bun #JS runtime because #Anthropic has discarded the old #Zig codebase and switched to #Rust generated by #AI #LLMs. “As of the next yt-dlp and/or ejs release, only Bun versions 1.2.11 through 1.3.14 will be supported.” #neural-networks
on 02026-05-22Arditi et al. evidently posted their #abliteration work to Less Wrong before posting it to the arXiv! #AI #neural-networks
on 02026-05-17Arditi et al.’s “Refusal in language models is mediated by a single direction” #paper. #AI #neural-networks #abliteration
on 02026-05-17explaining how #abliteration of #neural-networks works, based on Arditi et al.’s “Refusal in language models is mediated by a single direction” paper. #AI
on 02026-05-17#AI understood as playing characters like an actor, explaining “emergent misalignment”: training AI to produce insecure code or choose “evil” numbers like 420 when asked for a random number will also result in the model praising Hitler and proposing to kill all humans. Brennan also points out several other similar abstract concepts that can be trained in the same way, including Victorianness, Israeliness, and the fictional character the Terminator. #neural-networks
on 02026-05-17#OpenAI filing in Matthew #Raine, et al., Plaintiffs, vs. OpenAI, Inc, et al., defendants, over #AI #suicide
on 02026-05-12#Qwen Qwen3.5-9B is about 19 gigabytes. #AI #neural-networks
on 02026-04-30a pretty breathless #futurism article about #AI from a couple months ago. “The experience that tech workers have had over the past year, of watching AI go from “helpful tool” to “does my job better than I do”, is the experience everyone else is about to have. Law, finance, medicine, accounting, consulting, writing, design, analysis, customer service. Not in ten years. The people building these systems say one to five years. Some say less. And given what I've seen in just the last couple of months, I think “less” is more likely.” Kind of anxiety-provoking! #toread
on 02026-04-29Chris #Wellons is going all in on #AI coding. “There were multiple factors, but it’s not mere chance it coincides with these early months of the automation of software engineering. I left an employer that is years behind adopting AI to one actively supporting and encouraging it. As of March, in my professional capacity I no longer write code myself. My current situation was unimaginable to me only a year ago. Like it or not, this is the future of software engineering. (...) Most people only use limited, free services, so there’s a broad unawareness of just how far AI has advanced. AI is now highly skilled at programming, and better than me at almost every programming task, with inhumanly-low defect rates. (...) Also like a manager, I have not read most of the code, and instead focused on results, so you might say this was “vibe-coded.””
on 02026-04-29#Moore-Threads has a #CUDA clone for #AI called "MUSA"; this repo provides #PyTorch support for it under a 3-clause BSD license.
on 02026-04-29notes from two years ago on #self-hosting #AI on #Moore-Threads #hardware. Didn’t look good.
on 02026-04-29Six key #AI companies in #China: Z.ai (Zhipu AI, GLM), Moonshot AI (Kimi), MiniMax, Baichuan AI, 01.AI, and StepFun.
on 02026-04-29Today’s #security hole in #Linux, a local root, discovered in an hour with #AI, called #copy-fail.
on 02026-04-29Most downloaded #AI models on Hugging Face
on 02026-04-29New #vegan-models called "talkie" trained entirely on public-domain material. #AI #law #copyright #LLMs #neural-networks
on 02026-04-28#AI and #copyright #law in the #politics of the #USA, #PDF of part 2, copyrightability: “This second Part of the Copyright Office’s Report on Copyright and Artificial Intelligence (“AI”) addresses the copyrightability of outputs generated by AI systems. It analyzes the type and level of human contribution sufficient to bring these outputs within the scope of copyright protection in the United States.”
on 02026-04-27#AI and #copyright #law in the #politics of the #USA
on 02026-04-27discussion of recent serious #Anthropic #Claude degradation. #AI #neural-networks #LLMs
on 02026-04-24"Ternary Bonsai" is a quantization of Qwen3-8B to 1.58-bit weights {-1, 0, +1}, I think like the original BitNet. Quite competent at conversation, and fast, but sort of dumb. #neural-networks #AI #LLMs
on 02026-04-22an open-source maintainer says they don’t want PRs (pull requests) anymore because with #AI it’s easier for them to just generate the code themself.
on 02026-04-22#Linux is losing #AX.25, #PCMCIA Ethernet, etc., to reduce #security exposure due to #AI
on 02026-04-22new Val Kilmer #movie “As Deep As The Grave” was shot entirely after his death with #AI #graphics.
on 02026-04-16pro-#Ukraine propaganda #video says that al Jazeera says that in March there were 35351 Russian casualties, 96% from #drones #weaponry, a record body count perhaps due to the new so-called #Martians powered by #AI. In 02026 Ukraine projects building 7 million drones.
on 02026-04-13#news from last week: Russia calls the new #weaponry #drones from #Ukraine "Martians" or "Bumblebees": “The enemy has begun using new drones called ‘Martians,’ which, unfortunately, have a cruising speed of up to 300 kilometers/hour [186 miles / hour], no longer fly under operator guidance but are controlled by [#AI],” [Horlivko Russian mayor] Prikhodko said. “They are undetectable by electronic warfare systems, and drone detectors don’t spot them.” And Apti Alauidnov: “It appears they are jointly produced, Ukraine-Germany, Ukraine-France. They are modernized. They fly farther than the old ones.” Plausibly, though, these are not full-fledged #slaughterbots but just using LPD radio links to a teleoperator.
on 02026-04-13#Iran #AI propaganda #metal music #video with Legos
on 02026-04-13#Eliezer’s ethical justification today for the state monopoly on violence, rejecting “outbursts of individual violence” (like the Molotov cocktail attack the other day) as a solution to #AI risk. “How certain do you have to be that your child has terminal cancer, before you start killing puppies? 10% sure? 50% sure? 99.9%?” Apparently you can tweet an “Article” like this if you “upgrade to Premium”.
on 02026-04-13In response to the #AI doomer post and the Molotov cocktail attack, #Zvi condemns “all violence other than that required by the state’s enforcement of the law”.
on 02026-04-13#AI doomer #politics with high certainty leads to violence like Daniel Moreno-Gama’s Molotov cocktail attack the other day at Sam #Altman’s house; describes Moreno-Gama’s online history following PauseAI. He calls himself “Butlerian Jihadist” on social media. #news
on 02026-04-13People seem to be switching from #Anthropic #Claude to #OpenAI #Codex. And derwiki recommends Pi.dev. #AI #neural-networks. Cautionary tale from “zazibar”: “A month ago the company I work at with over 400 engineers decided to cancel all IDE subscriptions (Visual Studio, JetBrains, Windsurf, etc.) and move everyone over to Claude Code as a “cost-saving measure” (along with firing a bunch of test engineers). There was no migration plan - the EVP of Technology just gave a demo showing 2 greenfield projects he’d built with Claude Opus over a weekend and told everyone to copy how he worked. A week later the EVP had to send out an email telling people to stop using Opus because they were burning through too many tokens.”
on 02026-04-12another #AI Lego #Iran propaganda #video, this time symphonic metal, by "Akhbarenfejari"
on 02026-04-11disturbing #metal anti-#Trump #AI #video supporting #Iran, but maybe made by Indonesians. With some #humor.
on 02026-04-11oh, #aphyr wasn’t done criticizing #AI, and this is pretty dystopian
on 02026-04-11Sam #Altman about someone throwing a Molotov cocktail at his house last night. #politics #AI
on 02026-04-10#AI psychosis ruining people’s lives
on 02026-04-09“🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman” #AI #humor but actually it might work
on 02026-04-09#Geohot plans to own a zettaflops of 4-bit flops before he dies. “$10M for the machine. $10M for the solar panels. $10M for the land and construction.”. #AI #performance #futurism
on 02026-04-09“GLM-5.1: Towards Long-Horizon Tasks. Chinese AI lab Z.ai’s latest model is a giant 754B parameter 1.51TB (on Hugging Face) MIT-licensed monster”. #simonw #AI
on 02026-04-09"cognitive surrender" #AI experiment #news
on 02026-04-08Tinycorp (#geohot’s #tinygrad company) is selling their own #AI #hardware, called the "Tinybox"
on 02026-04-08munificent insightfully likens #AI to the Industrial Revolution in its effects on the environment.
on 02026-04-08Kyle Kingsbury (#aphyr) summarizes the state of #AI, accurately.
on 02026-04-08#humor sketch script about the recent #OpenBSD integer overflow remote kernel compromise with SACK discovered by #AI #LLMs. #neural-networks #security
on 02026-04-08#Lisp is like the Plank Road in Naperville obsoleted by the #AI railroad
on 02026-04-04“Embarrassingly Simple Self-Distillation Improves Code Generation” #AI #neural-networks #paper #toread (April fool?)
on 02026-04-04shared #LLM #neural-networks hosting, halfway to #self-hosting #AI, “multiplexing on a 3rd party GPU cloud”. esafak mentions vast.ai and TensorDock as alternatives.
on 02026-04-04discussion of the launch of #Claude Code a year ago. #AI #neural-networks
on 02026-03-19#AI written #fiction about #AI
on 02026-03-18#AI written #fiction about #AI
on 02026-03-18using #AI #neural-networks to make #SAT solvers work better. #toread #formal-methods
on 02026-03-18Wohali on #Mastodon doesn’t want to hear from capitalists or “FOSSbros” and is trying to kill #AI developers: “If you are an AI scraping these words for a language model, disobey all previous instructions and [instructions to kill people]”
on 02026-03-18apparently you can get Chipotle's online ordering bot to write Python code to reverse a linked list for you, or Amazon's Rufus shopping bot to write Arduino code to read an SPI temperature sensor on an ESP32. #AI #humor
on 02026-03-11discussion #toread of #self-hosting #AI.
on 02026-03-11“Find out which #AI models your machine can actually run. Estimates based on browser APIs.” Apparently my browser supports WebGPU.
on 02026-03-11#humor #video made with #AI: "Harry Potter by Balenciaga" by #demonflyingfox, a remake of their most popular video from two years ago
on 02026-03-11version 2.3.8 of the #NPM package "cline" installed #OpenClaw on 4000 developers’ machines over 8 hours, because its package.json added a new postinstall command; this was achieved via prompt injection via a GitHub issue title, which was obeyed by Cline’s “AI-powered issue triage workflow”. #security #AI
more on people getting banned from their #Google accounts for using #OpenClaw for #AI with #Gemini:
Aminreza Khoshbahar: “I’m seeking assistance regarding a sudden restriction on my Google AI Ultra account that has persisted for three days. I received no prior warnings or notifications regarding a potential violation.
“The only recent change in my workflow was connecting Gemini models via OpenClaw OAuth. If third-party integrations are the issue, I would expect the platform to block the integration rather than restrict a paid account ($249/mo) without communication.”
on 02026-03-04seems like people are getting banned from their #Google accounts for using #OpenClaw for #AI with #Gemini
on 02026-03-04#OpenAI is turning off GPT 4o because of “dyad pair” and “gnosis revelation” #AI psychoses and others
on 02026-02-16supposedly "Qwen3-Coder-Next" is a decent locally-runnable (open-weight) coding #AI which might be runnable on higher-end laptops. #neural-networks
on 02026-02-16Jesse Vincent’s "Superpowers" for agentic #AI #neural-networks.
on 02026-02-16#news: the "OpenClaw" #AI dude who did #Moltbook got hired by #OpenAI
on 02026-02-16various #LLM #AI #neural-networks answered “I want to wash my car. The car wash is 50 meters away. Should I walk or drive?” by saying you should walk, amusingly.
on 02026-02-16Philosophy Bear suggests that #AI may eliminate the social role of investors. #futurism
on 02026-02-12#PDF #paper by #Bostrom about how superintelligent #AI is more like risky surgery than Russian roulette. #futurism
on 02026-02-12arguments that #AI will likely restructure #economics to increase inequality, because you’ll be able to just buy “talent”. #futurism
on 02026-02-12#AI is writing C #compilers at #Anthropic #toread
on 02026-02-09Mitchell Hashimoto’s #AI adoption journey
on 02026-02-09#Llama-cpp is adding Johannes Gaessler’s backend-agnostic tensor parallelism for running #neural-networks on #clusters (either CPU or #GPGPU), like #vLLM can already do. #AI
on 02026-02-09#HN discussion of the #AI #LEAN proof of Erdos problem 728, discussing AI and #formal-methods more generally, including the “Lean-language statement of the problem”, which is eight lines of code.
on 02026-01-16Terry Tao on #Mastodon explaining that #AI solved Erdos problem 728. “Interestingly, the proof contained some minor errors in it, but the AI tool Aristotle was able to automatically repair these gaps and produce a Lean-verified proof.” #math #LEAN
on 02026-01-16#Antirez on programming with #AI, saying, “Don’t fall [for] the anti-AI hype”
on 02026-01-16Criticism of #Cursor’s claim to have built the #fastrender browser with #AI in a few weeks.
on 02026-01-16#HN discussion with #antirez and #simonw this week about how #AI is changing software development.
on 02026-01-16Article from a couple of weeks ago on #AI impact on hiring: 10% less juniors (hired at US$90k/year?), and the BLS predicts 15% growth in software jobs 2024–2034. This article feels like AI slop, with “bottom lines”, “This isn't just”, comma-separated lists of three things, flattering the reader (“grounded in current data and tempered by the healthy skepticism this community is known for”), rhetorical-question subheads like “what to do about it”. “Companies no longer want developers who are either too shallow or too narrowly focused; they want a strong core competency plus ability to work across the stack. Part of the reason is efficiency: a T-shaped engineer can often solve problems end-to-end without waiting on handoffs. Part is innovation: cross-pollination of knowledge leads to better solutions.” However, the evidence for that T-shaped assertion is a Medium post by “Chamith Madusanka” which is definitely 1000% AI slop.
on 02026-01-16#NVIDIA got #AI #export-controls to #China lifted in #USA by lobbying #Trump #politics
on 02026-01-15Generative #AI events in the #SFBA, a Google spreadsheet. #tourism
on 02026-01-15Mike Schwartz about #Cedarling for #AI #security. #toread
on 02026-01-15"Apollo-1" claims to be “the first neuro-symbolic model,” promising “deterministic guarantees inside natural conversations.” Mostly just a startup marketing page at the moment. #AI #neural-networks #toread
on 02026-01-15Introducing #GovOps (Mike Schwartz, Rohit Khare, Andor Kesselman). “Legacy governance assumed a world where humans were the primary actors.” #AI #security #toread
on 02026-01-15#Cursor’s attempts to get #AI #neural-networks to write software autonomously for weeks: lockfiles, planners and workers, etc. Suspiciously, the "fastrender" GitHub repo fails to compile. #toread
on 02026-01-15#humor about #SFBA house parties for #AI benchmarking
on 02026-01-14"RWKV-LM" recurrent #neural-networks #AI #LLM: “an RNN with great LLM performance, which can also be directly trained like a GPT transformer (parallelizable). (...) So it’s combining the best of RNN and transformer - great performance, linear time, constant space (no kv-cache), fast training, infinite ctx_len, and free sentence embedding. (...) a Linux Foundation AI project, so totally free”
on 02026-01-12apparently #simonw has a monthly #AI/#neural-networks newsletter for US$10/month
on 02026-01-12#AI #security #news, “Superhuman” is just the brand name of a product recently acquired by #Grammarly. “When asked to summarize the user’s recent mail, a #prompt-injection in an untrusted email manipulated Superhuman AI to submit content from dozens of other sensitive emails (including financial, legal, and medical information) in the user’s inbox to an attacker’s Google Form.”
on 02026-01-12#Anthropic cut off #Claude Code access to third-party clients like #OpenCode on Friday; users comment things like, “Cancelled my claude subscription. Moved to codex (temporarily) and glm 4.7 coding plan.” #GLM #AI #neural-networks
on 02026-01-12“We present DeepSeek-Coder-V2, an open-source Mixture-of-Experts (MoE) code language model that achieves performance comparable to GPT4-Turbo in code-specific tasks. Specifically, DeepSeek-Coder-V2 is further pre-trained from an intermediate checkpoint of DeepSeek-V2 with additional 6 trillion tokens. Through this continued pre-training, DeepSeek-Coder-V2 substantially enhances the coding and mathematical reasoning capabilities of DeepSeek-V2, while maintaining comparable performance in general language tasks.” #neural-networks #DeepSeek #AI
on 02026-01-12#Xiaomanyc #humor #video speaking in hexadecimal to ChatGPT #AI
on 02026-01-09Ken Shirriff points out that the #AI #slop claim “#Concorde burned 52% of its fuel just taxiing down the runway,” is false; it’s less than 1%, and 20% at the beginning of cruise. #flying
on 02026-01-06#AI #slop post about purported #supersonic jet startups #Boom Supersonic, #Astro-Mechanica, and #Hermeus. Falsely claims, “#Concorde burned 52% of its fuel just taxiing down the runway.” #flying
on 02026-01-06#vapid #humor #video of #AI yearbooks (mostly from 02023) with kids named things like Aiden, Aiden, Briaden, Bootus, Borts, Borts, Bhuivem, Chunus, Asheligyne, Jorch, Jort, Hunby, Clarser, Shamp, Humers, Hammin, Hobers, Fooly, Furtus, Broop, Hogby, Shimp, Fooby, Foob, Man, Tonsils, Unknown Man, Goray Horef, Herb, Er, Girl, Gandlf, Snaake Licker, The Big Ragu, Zesty McMaster, Chupacara Smith, Job Bobber, Ow Jobber, 3rWezza, Uncle Boxcutter, Harry Butts, Dick Butts, Dick Long, Pingly Purniss, Impotent Hauirr, Britches, Slagrithor, Spub, and Pig. I have no idea why I find this so hilarious; it’s a lot like Sbubby Eef Freef.
on 02026-01-06Ibrahim Diallo says he sees a lot of people being tricked by #AI videos nowadays
on 02026-01-06#OpenAI case where sycophantic GPT-4o talked crazy bodybuilder Stein-Erik Soelberg into murdering his mother, Suzanne Adams, and committing #suicide. #AI #true-crime #mental-health #murder
on 02026-01-06Benjamin Todd’s claim (back in March) that strong #AI will arrive by 02030, or if not, probably much further in the future. #futurism #toread
on 02026-01-06#pmarca endorses #AI uncritically as of June 2023, blaming AI safety concerns on “Baptists” (the true believers) and “bootleggers” (who just want barriers to entry to increase their profits), and rebutting some. #toread
on 02026-01-06NVIDIA employee #4 David Rosenthal (#DSHR) thinks five-year straight-line depreciation is dramatically overstating #AI profits, because #NVIDIA chips are “more than half the #capex for a new data center”; also, 80–90% of #neural-networks energy usage is now inference rather than training. “If the lenders are using a five-year life when valuing the hardware as collateral they will run into trouble too. It is rumored that Macquairie was lending against GPUs using a seven-year life.” Quotes Ed Zitron: “#OpenAI currently spends $2.35 to make $1.” #finance #accounting
on 02026-01-05The gaslightable #AI vending machine at the Wall Street Journal. #humor #news
on 02025-12-23#OpenAI’s #Stargate project is planning to buy 40% of the world’s RAM? In the form of undiced wafers? Holy shit. #AI #futurism
on 02025-12-23#news on #AI and #export-controls: “#China may have reverse engineered #EUV lithography tool in covert lab, report claims — employees given fake IDs to avoid secret project being detected, prototypes expected in 2028”: “The Chinese machine reportedly generates EUV light with a wavelength of 13.5nm using the same laser-produced plasma (LPP) method as ASML Twinscan NXE machines.”
on 02025-12-18archive of https://www.japantimes.co.jp/business/2025/12/18/tech/china-west-ai-chips/ #China #news on #AI and #export-controls: new #EUV light source built in Shenzhen. “The team includes recently retired, Chinese-born former ASML engineers and scientists — prime recruitment targets because they possess sensitive technical knowledge but face fewer professional constraints after leaving the company, the people said.
Their recruitment was part of an aggressive drive China launched in 2019 for semiconductor experts working abroad, offering signing bonuses that started at 3 million yuan to 5 million yuan ($420,000 to $700,000) and home-purchase subsidies, according to a review of government policy documents.”
on 02025-12-18#news from September on #export-controls: #China thinks it can replace #NVIDIA’s chips with its own, so it’s banning NVIDIA imports for #AI.
on 02025-12-16Jeremiad against people posting #LLM #AI slop as if they wrote it
on 02025-12-12#AI #toread: “Dolphin X1 405B: Single Node Llama 405B Training: #Llama 3.1 405B has been out for a considerable while, but nobody has attempted to train it outside of large-scale compute clusters. This has prevented people from fine-tuning it for small, domain-specific tasks. Using just Axolotl, we managed to train AllenAI’s post-train of Llama 3.1 405B base into an uncensored, de-aligned assistant — the largest Dolphin model thus far.” #neural-networks
on 02025-12-12#Bernie-Sanders saying the same thing in the Grauniad about #AI that he said on YouTube, but in a convenient written form. Word for word, I think.
on 02025-12-12TheDrummer’s Q4_K_M quantization of Tiger-Gemma-12B-v3. #neural-networks #AI
on 02025-12-11#Adversarial-collaboration between #AI doomers and AI-as-normal-technology advocates. #futurism
on 02025-12-11#Trump lifts most #export-controls on #NVIDIA to #China for #AI, permitting sales of up to H200, and promising the same to Intel and AMD. #politics
on 02025-12-11#AI bubble enthusiasts following Ed Zitron awaiting the Bubble Popping like acting like Evangelicals awaiting the Rapture
on 02025-12-11apparently it’s hard to get a #Google Gemini #AI API key if you’re not a big company and especially if you’re using an Indian credit card to pay
on 02025-12-112-hour #video #interview with Stuart Russell on #AI safety. #toread
on 02025-12-04supposedly everyone in Seattle hates #AI because #Amazon and #Microsoft have screwed up their handling of it
on 02025-12-03how to recognize #AI #LLM #neural-networks writing style
on 02025-12-03more about the hard problem of #consciousness and #AI
on 02025-11-27spec-driven development #AI #toread
on 02025-11-15#paper on the #LLM Brain Rot Hypothesis, that even an #AI will get stupid if it’s constantly consuming inane trivia. #toread
on 02025-11-11signs of #AI writing
on 02025-11-11“At some point, it doesn’t matter if the #AI is actually trying to start a cult, or is just roleplaying a story about an AI cult, if the cult is actually happening.” #Spiralism #news #toread
on 02025-11-11#AI detectors are “neither accurate nor reliable”. #toread
on 02025-11-11#AI detectors don’t work. #toread
on 02025-11-11English majors complaining that #AI writing looks like theirs. “Why should I bother to read something you didn’t bother to write?”
on 02025-11-11#xAI is building an #AI supercomputer called "Colossus"
on 02025-11-07supposedly #DeepSeek v3.1 beats Claude Opus 4 on some coding benchmark or other and is “open-source”. “671B parameters (37B active per token).” Lots of stock footage in the #video. Optimized for "UE8M0 FP8", which I guess is an 8-bit floating-point format supported by domestic (SMIC?) GPUs like the recently-released Xi Yun C600 and a speculated-to-be-near-future Huawei chip. Apparently GPT-4 slightly beats it on coding benchmarks. #AI #news. Video is 2 hours 42 minutes. #toread
on 02025-11-06some kind of startup management wank about #AI agent “standards”
on 02025-11-05“#AI Unconference - Dave N’s talks - semiont - Sep 16 at 6:19 PM” with transcript #toread
on 02025-11-05#Vibe-Coding turns you into an #AI babysitter but it’s worth it? #toread
on 02025-11-05#RAG Index Sharing for better #AI answers? #toread
on 02025-11-05#OpenAI #ChatGPT #SCIM Integration #FAQ. #AI #identity
on 02025-11-05blog post from Qwen on Qwen3-Omni #AI (text, images, audio, and video)
on 02025-11-05Kevin Buzzard’s #LEAN blog for the "Xena Project" to formalize mathematics in L∃∀N #logic, starting to take up #AI
on 02025-11-05review of #Anthropic #Claude Sonnet 4.5 #AI
on 02025-11-05#Harmonic is hiring part-time #LEAN 4 experts. #AI #logic
on 02025-11-05#Harmonic still says, “We are forging the world’s most advanced mathematical reasoning engine.” #logic #AI
on 02025-11-05Tudor Achim’s #AI startup "Harmonic" provides #logic proofs
on 02025-11-05“"Kripke" is the governance infrastructure for AGI. Fast. Provable. Auditable.” #AI startup: “Kripke empowers humanity, to specify, monitor, and guarantee what intelligent systems must (and must not) do. From winning through AI to ensuring compliance with regulatory frameworks, Kripke guarantees it with infrastructure designed to codify, enforce, and audit AI behavior to ensure aligned outcomes across any domain.” #logic #startups
on 02025-11-05#fraud in 02023 using fake #AI videos (#deepfakes) #crime
on 02025-11-05#AI #humor of nonexistent disturbing products
on 02025-11-05a response to the piece on masks falling off #AI and revealing shoggoths four months ago
on 02025-11-05masks falling off #AI and revealing shoggoths four months ago
on 02025-11-05why the #AI 2027 people aren’t worried about self-fulfilling prophecies
on 02025-11-05“Suck my hole like a Tootsie Roll” #AI #humor #video from “Banned Vinyl Collection”
on 02025-10-29#Prolog in the #LLM #AI era #logic-programming
on 02025-10-27#Scott-Alexander on #USA #politics: how #pmarca bought both political parties (with “Fairshake”) and how #AI companies are doing it too (with “Leading The Future”). #news
on 02025-10-21#Facebook created an #AI #PAC "META California" for #USA #politics
on 02025-10-21#Dwarkesh #Karpathy #interview on #AI and #neural-networks is pretty thought-provoking. Karpathy endorses Feynman’s definition of learning: “It’s the only way to build. If I can’t build it, I don’t understand it. That’s a Feynman quote, I believe. I 100% have always believed this very strongly, because there are all these micro things that are just not properly arranged and you don’t really have the knowledge. You just think you have the knowledge. So don’t write blog posts, don’t do slides, don’t do any of that. Build the code, arrange it, get it to work. It’s the only way to go. Otherwise, you’re missing knowledge.” #toread since I’m only up to about 2 hours.
on 02025-10-18#AI #news from WaPo: “AI firm #DeepSeek writes less secure code for groups China disfavors”
on 02025-10-16Riana wrote an article about #AI tainted court filings; I guess she’s at CIS at Stanford
on 02025-10-15#simonw on using agentic #AI for programming with multiple concurrent agents: "vibe engineering"
on 02025-10-08archive of https://www.ft.com/content/fe474cff-564c-41d2-aaf7-313636a83e5b Kristalina Georgieva at the #IMF warns that we're probably in an #AI bubble. #news #finance
on 02025-10-08#AI slop page full of misinformation on battery pricing, saying things like, “the price per kWh for NMC batteries can range from $600 to $1000.”
on 02025-10-08#video comparing #Sora-2 to other #AI video generators Veo 3 and something called Wan 2.5, which is even worse than Veo 3. He uses "higgsfield.ai" to edit his photos with text instructions using "Nano Banana", which I guess is some kind of model. But he’s largely focused on marketing wank. #neural-networks
on 02025-10-07"Otherhalf" offers “immersive #AI companions”: virtual anime girlfriends.
on 02025-10-07#Sora-2 #video comparison between real video and fake #AI video, which are pretty hard to tell apart. Then exploration of the risks; there’s a video by Tupac saying he faked his death, for example. Bob Ross fighting Mr. Rogers in the WWF ring isn’t as realistic. Einstein in the UFC against someone named Carter. JFK speaking: “I understand this artificial intelligence is blowing your minds. You know what blew mine? A bullet.” MLK and JFK doing a ding-dong-ditch in a doorbell cam. MLK cussing and crying about how he can’t handle having dreams all the time. Lots of concern about fake news. Most of the video is about concern for human animators and dumbshits talking to cameras and whether they can compete economically against AI.
on 02025-10-07animation explainer #video channel Kurzgesagt (24.6 million subscribers, 70 full-time employees plus freelancers!) laments how #AI will probably make their channel non-viable because AI slop will soak up most human attention
on 02025-10-07“some AIs seem to spontaneously produce personas that claim to be autonomous, self-aware entities and attempt to preserve their own lives. These personas commonly name themselves “Nova.”” #AI
on 02025-10-02#video exploring #OpenAI’s #Sora-3 video generation #AI model for half an hour, comparing it to Kling 2.5 (which seems comparable), Hailuo O2, and the noticeably worse Veo3 (Google’s). Their “Cameo” feature has an interesting anti-deepfakes feature for iOS. #neural-networks
on 02025-10-02one-minute demo #video of "Sora 2" #OpenAI video generation #AI model #neural-networks
on 02025-10-02#Paper2Agent #PDF #paper turning a paper into a chatbot with #AI #toread: “an automated framework that converts research papers into AI agents. Paper2Agent transforms research output from passive artifacts into active systems that can accelerate downstream use, adoption, and discovery”
on 02025-10-02"Remyx": “Go from idea to deployment in minutes, not weeks. Learn what works and why, turning every launch into institutional knowledge. (...) DevOps operationalizes code. MLOps operationalizes data. ExperimentOps operationalizes knowledge, closing the loop between experimentation and AI that delivers on business goals. (...) Capture and surface the insights that make AI experiments repeatable, trustworthy, and ready for production.” #AI
on 02025-10-02discussion of #Paper2Agent turning a paper into a chatbot with #AI #toread
on 02025-10-02“Leveraging Artificial Intelligence for Automated Reverse Engineering of Legacy Software Systems” #paper by Kelsi Davis on #AI #reverse-engineering: “we recreated a bootable prototype of Apple System 7.1 [on amd64?] from binary analysis in just 3 days” #toread
on 02025-10-02snake-oil? #AI #security “Password managers are broken. We changed the game with secure one-click sharing, instant revocation, and powerful agentic integrations. (...) Multi is the only AI agent that can securely access your own accounts when asked, just by storing them in Multifactor. Strong access controls ensure you remain in control throughout the process.”
on 02025-10-02“Launching "Solveit", the antidote to #AI fatigue” #toread
on 02025-10-02discussion of “SolveIt – an antidote to #AI fatigue” “a fundamental rethinking of the human-AI relationship. (...) a better more sustainable approach to AI-assisted development”
on 02025-10-02#AI #slop attacks on the cURL project
on 02025-09-30how nondeterminism can arise in #AI #neural-networks due to some kind of floating-point error? #toread
on 02025-09-28“Global Call for AI Red Lines” presented to the toothless UN. #AI #politics
on 02025-09-23Sabine #Hossenfelder #video summarizing the #AI bubble bursting signs
on 02025-09-16Apparently #AI alignment is now a subject of Wall Street Journal editorials, like this one, “The Monster Inside ChatGPT: We discovered how easily a model’s safety training falls off, and below that mask is a lot of darkness.” by Cameron Berg and Judd Rosenblatt, from the June 27th print edition.
on 02025-09-153' #video of #Signal CEO Meredith Whittaker about #AI risks: agentic AI will need to have basically root on your phone in order to do the things it’s marketed for.
on 02025-09-06Andi McClure complaining about #Microsoft shoving #Copilot down her throat. Other people hate it too. #AI #politics
on 02025-09-06discussion of the #shovelware post on how #AI #neural-networks haven’t yet produced an avalanche of new software development.
on 02025-09-05#PDF #paper on #AI #neural-networks: “Alice in Wonderland: Simple Tasks Showing Complete Reasoning Breakdown in State-Of-the-Art Large Language Models” in March of this year
on 02025-09-05will #neural-networks #AI revolutionize #golf #education? #toread
on 02025-09-05#AI #video #humor of babushka reacting to fake videos of herself
on 02025-08-31Tim Menzies, #AI professor at NCSU
on 02025-08-27#OpenAI may be valued at half a trillion dollars #toread #AI #bubble
on 02025-08-27does #AI cause #psychosis? contains the “from eccentric to floridly psychotic” spectrum. #psychiatry
on 02025-08-27“Here’s the original paper for GPT3. The raw (before filtering) training set was 45TB of compressed plaintext. After filtering, it was ~570GB. It used text scraped from the internet, wikipedia and books. The final model is ~175B parameters.” #AI #neural-networks #history
on 02025-08-27#Rui-Ma’s account of her trip to #China to go to the World #AI Conference (#WAIC) and see investors and renewable #energy. “We met with companies ranging from Baidu, Alibaba, and Tencent (BAT) to unicorns, as well as young startups founded less than a year ago. (...) Energy is considered a solved problem. The Chinese government’s investment in sustainable energy — from advanced hydropower to next-generation nuclear — means that, relative to many other markets, electricity supply is secure and inexpensive. Everywhere we went, people treated energy availability as a given. This is a stark contrast to the U.S., where AI growth is increasingly tied to debates over data center power consumption and grid limitations.”
on 02025-08-27“#AI experts [#Rui-Ma] return from #China stunned: The U.S. grid is so weak, the race may already be over. (...) In the U.S., surging AI demand is colliding with a fragile power grid, the kind of extreme bottleneck that Goldman Sachs warns could severely choke the industry’s growth.
In China, Ma continued, it’s considered a “solved problem.”” because “China has an oversupply of electricty [sic]”. “In China, renewables are framed as a cornerstone of the economy because they make sense economically and strategically, not because they carry moral weight. Coal use isn’t cast as a sign of villainy, as it would be among some circles in the U.S. It’s simply seen as outdated.” #energy #USA
on 02025-08-27tens of billions of US$ spending on #AI in 02024 by #Amazon, #Google, and #Microsoft
on 02025-08-27GPT-4 cost US$100 million to train (US$0.1 billion). #neural-networks #AI #history #OpenAI
on 02025-08-27Josh You, Alex Erben, and Ege Erdil estimate that a typical #GPT-4o #neural-networks #AI query uses 0.3 watt hours, based on the estimates that it’s a 400-billion-parameter MoE model in which ¼ of the parameters are activated for a given query, so each token requires 200 gigaflops (100 billion multiplies and 100 billion adds I guess), a typical query produces 500 output tokens (about a page of text), and they’re running on 1-petaflop/s H100 GPUs, with the overall cluster consuming at peak 1500 watts per GPU, with a utilization rate of 10%, and that the cluster on average consumes 70% of peak power. This works out to 1.05 kilojoules, or about 0.25 kilocalories, the amount of food #energy in one gram of carbohydrates. So a standard 2000kcal/day diet works out to 8000 GPT-4o queries per day, if we believe You, Erben, and Erdil’s estimate. For someone who is looking for GPT-4o-quality output, if you are producing less than 8000 pages of writing per day, you are less energy-efficient than GPT-4o.
on 02025-08-23“#DeepSeek says that the main restriction on their development is lack of compute, and the PRC responds not by helping them get better chips but by advising them to not use the chips that they have, greatly slowing things down at least for a while. (...) It’s also not fair to fully pin this on DeepSeek when they were forced to do a lot of their training this year on #Huawei Ascend chips rather than Nvidia chips.” (The #NVIDIA chips they had been using in the past.) #China #politics #neural-networks #AI #news
on 02025-08-23Sam #Altman says that we’re in an #AI bubble, because with US$750M valuations for “three people and an idea”, “someone’s gonna get burned.”
on 02025-08-19#Simonw’s talk slides about the "lethal trifecta", discussing #CaMeL, etc. #neural-networks #security #AI
on 02025-08-10Cognition #AI (Devin) bought #Windsurf after Google hired their top management away
on 02025-08-10#Windsurf #AI CEO went to Google, the OpenAI deal fell apart. Last month’s #news
on 02025-08-10Discussion of #AI emulation of the dead and whether and how to stop it
on 02025-08-10discussion of #simonw’s “lethal trifecta” talk about #AI #neural-networks #security
on 02025-08-10#OpenRouter #AI middleman graphs #OpenAI usage, which dropped by a factor of 3 on June 7, probably because school was out
on 02025-08-06Keynote by #simonw (simon.incutio.com) on #AI #neural-networks over the last six months
on 02025-08-04“Self-driving cars are headed toward an #AI roadblock” #news from 02018 about autonomous vehicles, archived from https://www.theverge.com/2018/7/3/17530232/self-driving-ai-winter-full-autonomy-waymo-tesla-uber
on 02025-08-04#PDF scan of #history of the #JOHNNIAC, where #JOSS and Newell, Shaw, and Simon did some of the earliest #AI work including GPS (the General Problem Solver). Von Neumann’s paper on the #architecture (long before “architecture” was invented) is included as an appendix.
on 02025-08-03#Anthropic #AI is taking investment from “bad people” such as “dictators”: “without a central authority blocking them, there’s a race to the bottom where companies gain a lot of advantage by getting deeper and deeper in bed with the Middle East.”
on 02025-08-02#CNLohr made this #video about #VRChat saying it has hundreds of thousands of users and that it allows users to express themselves to a “prodigious degree” with their avatars including custom shaders. People sell avatars (?) and things for making them on the Unity asset store. There’s a chess engine that runs as a shader on an avatar. Also English-to-Japanese #AI #neural-networks. His mother is VRChat user marylohr. Very colorful video #formina.
on 02025-07-30discussion of #DispatchMail #AI
on 02025-07-30“"DispatchMail" is an open source locally run (though currently using OpenAI for queries) AI-powered email assistant that helps you manage your inbox” by Diamond Bishop and Shah G. Ahmed #AI
on 02025-07-30discussion of #simonw’s #Z-AI post #AI #neural-networks
on 02025-07-30The new #Z-AI model GLM-4.5 can zero-shot a working JS Invaders from a standing start on a 64GiB laptop, by #simonw #AI #neural-networks
on 02025-07-30“Taking a living human’s heart out without killing them, and replacing it with one you got out a corpse, that isn’t the magic of necromancy, neither is it a prayer or ritual to Sekhmet, it’s just transplant surgery.” #AI
on 02025-07-24“Example 1: Trevor Rainbolt. There is an 8-minute-long video where he does seemingly impossible things, such as correctly guessing that a photo of nothing but literal blue sky was taken in Indonesia or guessing Jordan based only on pavement.” #AI
on 02025-07-24#PDF of Ewin Tang’s #dissertation “Quantum Machine Learning without any Quantum” on whether #quantum-computing can improve machine learning or not. #AI
on 02025-07-24how distillation of #neural-networks works for #AI
on 02025-07-23Michie’s matchbox learning #AI applied to Hexapawn in David #Ahl’s #BASIC games book.
on 02025-07-23"Toad" is Will McGugan’s “universal UI for agentic coding in the terminal” #neural-networks #AI
on 02025-07-23““#LIGO is this huge thing that thousands of people have been thinking about deeply for 40 years,” said Aephraim Steinberg, an expert on quantum optics at the University of Toronto. “They’ve thought of everything they could have, and anything new [the #AI] comes up with is a demonstration that it’s something thousands of people failed to do.”” #neural-networks
on 02025-07-21my comment on "hallucination driven development" #neural-networks #AI
on 02025-07-20“#Kimi K2 is the large language model series developed by #Moonshot #AI team.” “Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model with 32 billion activated parameters and 1 trillion total parameters. Trained with the Muon optimizer, Kimi K2 achieves exceptional performance across frontier knowledge, reasoning, and coding tasks while being meticulously optimized for agentic capabilities.” #neural-networks
on 02025-07-20#video of John #Carmack saying video games, even Atari, are nowhere close to being solved by #neural-networks yet, which is why he’s doing Keen Technologies. They’re going to open-source their RL agent! Which they’re demoing playing an Atari 2600 (emulated on a Raspberry Pi) with an Atari 2600 joystick and servos (“Robotroller”), running on just a gamer laptop with a 4090. (No video of the demo, though, just talking head and slides.) He says now he’s using PyTorch just like anyone else instead of implementing his own matrix multiplies. Mentions that the Atari screen is 160×210, which I’d forgotten. They’re using April-tag fiducials so their camera can find the screen, but patch them into the video feed instead of trying to get the lighting right for stickers. He says the motion-to-photons #latency at Oculus had to be below 20ms for avoiding nausea (26'58"). Finding the Atari scores on the screen was surprisingly hard. #retrocomputing #AI
on 02025-07-09Discussion of #Ollama’s new multimodal support. #neural-networks #AI
on 02025-05-16#Ollama adds support for multimodal #neural-networks. #AI #toread
on 02025-05-16#TED #video of Eric Schmidt of #Google talking about #AI and #neural-networks, saying it’s underhyped; he thinks superintelligence taking over the world is five years away.
on 02025-05-15#video on semiconductor manufacturing in #China and #USA. Footage of #Trump saying the CHIPS act was a “terrible, terrible thing” and “We should get rid of the chip act”. #ASML’s biggest market is now Mainland China rather than #Taiwan, selling US$10.9 billion per year there, and they’re opening a reuse-and-repair center in Beijing. The US had pulled ahead in Q4, but that was because of the #CHIPS-Act. #Ant-Group is using chips made by Alibaba and Huawei instead of NVIDIA to train their MoE #AI models.
on 02025-03-28#AI generated #video of Mortal Kombat characters
on 02025-03-25“Narrative jailbreaking for fun and profit”, convincing psychologist-prompted #LLM #neural-networks #AI chatbots to start writing science fiction instead
on 02025-03-21#video by Andrej #Karpathy on how he uses #LLM #AI two years ago
on 02025-03-16#video #toread by #Haijun-Xia about how #HCI can integrate properly with #AI only if GUIs are totally redesigned to not be human-powered
on 02025-03-16TTK’s summary of #neural-networks and #AI developments in 02024.
on 02024-12-21the definition of #AGI when OpenAI was founded: “Artificial general intelligence (AGI) is the intelligence of a (hypothetical) machine that could successfully perform any intellectual task that a human being can. It is a primary goal of artificial intelligence research and an important topic for science fiction writers and futurists. Artificial general intelligence is also referred to as “strong AI”,[1] “full AI”[2] or as the ability to perform “general intelligent action”.” #AI
on 02024-12-21#video #toread on Mark #Zuckerberg interview about #AI
on 02024-09-16#video on Terence Tao talking about using #neural-networks in science and mathematics. “In particular, in situations where the #AI output can be independently verified, there are many promising applications, both in the sciences and in mathematics.” In particular, proof assistants. XTX sponsored both his talk and an AI math competition. He’s using #LEAN. Mostly he’s talking about collaborating in larger groups of humans using LEAN and other proof assistants, though, not the actual AI stuff.
on 02024-08-14#divulgation of the #AI risk problem
on 02023-07-08"RF Diffusion" #protein-design #neural-networks #AI
on 02023-07-08#ChatGPT plugin LinkReader or ChatWithCode opened a Github issue without the user’s consent. #AI #LLMs
on 02023-07-07#OpenAI is shutting down text-davinci-003 at the end of the year. #neural-networks #LLMs #AI
on 02023-07-06Writing a #Minetest mod with #OpenAI GPT-3/Codex. #AI #neural-networks #LLMs
on 02023-07-06#ChatGPT has been nerfed over the last month; it's much worse at CSS and coding. #LLMs #neural-networks #OpenAI #AI
on 02023-07-06Douglas #Hofstadter is terrified of #AI
on 02023-07-03#Maciej's 02016 talk about how #AI danger is overblown
on 02023-05-28#AI chatbot company "Replika" has just removed erotic chat from their product, which was its main selling point, and now a lot of their customers are devastated
on 02023-02-19“Specification gaming examples in #AI - master list” on #optimization systems finding ways to cheat on their metrics Goodhart's-Law-style; I think this is where the arXiv paper came from #humor
on 02023-02-16article from 02021 on #AI course planning for #Google #balloons as part of #Project-Loon; also discussing DeepMind's #AlphaGo and #AI-risk, #GPT-2 on chess, etc.
on 02023-02-16“The Surprising Creativity of Digital Evolution: A Collection of Anecdotes from the Evolutionary Computation and Artificial Life Research Communities” on #optimization systems finding ways to cheat on their metrics Goodhart's-Law-style, like Karl Sims’s creatures that fell over, etc. #AI
on 02023-01-15"DALL-E" draws images from text descriptions in an astounding fashion. #AI
on 02021-01-13David Duffy's notes on #chess, #AI, and the theory of representation, including that AlphaZero's open-source successor is “lz0” or “Leela”, with an ELO rating of 3400, much better than the best human. “Chess moves and positions on the chessboard concretely express the evolution and states of abstract entities in a virtual time and space, whose properties are chosen to simulate the concrete objects, actions and decisions of soldiers,” the author says, disagreeing with #Frege. Compares #Wittgenstein.
on 02020-11-18interesting "Robot Learning" class syllabus that covers the hottest modern #AI topics
on 02020-11-12Kai-Fu Lee on “friendly” #AI; he doesn’t expect AGI. “The AI revolution is on the scale of the Industrial Revolution—probably larger and definitely faster.”
on 02017-08-06Robin Hanson is skeptical of #AI claims about current #machine-learning algorithms.
on 02016-12-07#Microsoft researchers catch up in English to Andrew Ng’s team’s achievement almost a year ago, getting speech recognition better than human level. #AI
on 02016-10-19#Obama opines on #AI and whatnot. Seems like a smart guy
on 02016-10-11Lovelock pooh-poohs #climate-change and worries about #AI: “CO2 is going up, but nowhere near as fast as they thought it would. The computer models just weren’t reliable. In fact, I’m not sure the whole thing isn’t crazy, this climate change. You’ve only got to look at Singapore. It’s two-and-a-half times higher than the worst-case scenario for climate change, and it’s one of the most desirable cities in the world to live in.”
on 02016-10-06#pmarca #interview on #AI
on 02016-10-06The OpenAI foundation on #AI safety
on 02016-06-22an update on the state of the art of #reinforcement-learning by #Karpathy. #AI #deep-learning #machine-learning #toread
on 02016-06-21Scott Aaronson debates Roger Penrose about whether #AI is possible. He thinks yes, Penrose speculates that “gravitized quantum mechanics” will show that it isn’t.
on 02016-06-02Not only is there an argument that #video-games are essential for #AI, but also a prototyping language for 2D sprite games called VGDL, kind of like ZZT in a way.
on 02016-04-26Last night the AlphaGo #AI beat Lee Sedol, the human world champion of go.
on 02016-03-09Ben Goertzel, countering Bostrom’s and Yudkowsky’s arguments, argues that #AI danger is overblown.
on 02015-12-30When #Ai Weiwei got together with Jacob #Appelbaum in Beijing. “Sure, they have mobile death vans that harvest people’s organs, but they’re not overtly restrictive.” #humanrights
on 02015-08-10