AINews Portal

News

50 items (filtered)

The raw AI news stream — everything we ingest across RSS, Hacker News, arXiv, GitHub, Bluesky, YouTube and more, newest first.

SourceAgeTitleSummaryStatus
bluesky
bluesky-ai
2mo"You are not seeing someone in a comic book kind of way say, ‘Hey, I want to build a biological weapon to kill everybody,’” Jacob Klein, th…"You are not seeing someone in a comic book kind of way say, ‘Hey, I want to build a biological weapon to kill everybody,’” Jacob Klei…
↗Not briefed
bluesky
bluesky-ai
2moTrain a 3.8B LLM for $998, hits 0.384 CORE 🤯 Hugo Vergnes trained a 3.8B LLM to 0.384 CORE on 65B tokens for just $998. This significantl…Train a 3.8B LLM for $998, hits 0.384 CORE 🤯 Hugo Vergnes trained a 3.8B LLM to 0.384 CORE on 65B tokens for just $998. This signifi…
↗Not briefed
hn
hn-ai
2moPCBGolf challenge: Are you better than an autorouter?—
↗Not briefed
bluesky
bluesky-ai
2moThe @hrcberkeley.bsky.social has released a human rights assessment of the use of large language models (LLMs) in the fields of law, journa…The @hrcberkeley.bsky.social has released a human rights assessment of the use of large language models (LLMs) in the fields of law, j…
↗Not briefed
bluesky
bluesky-ai
2moIt is just not that hard to build really bad bioweapons, but I guess they don't teach that in Cambridge math. Coxon also brags that OpenAI…It is just not that hard to build really bad bioweapons, but I guess they don't teach that in Cambridge math. Coxon also brags that O…
↗Not briefed
bluesky
bluesky-ai
2moDon’t be fooled. Congress isn’t cracking down on OpenAI.Don’t be fooled. Congress isn’t cracking down on OpenAI.
↗Not briefed
bluesky
bluesky-ai
2moDeepSeek V4.1 Flash tech report section on reward hacking. I wonder what repercussion signal means here, I don't see it mentioned anywhere …DeepSeek V4.1 Flash tech report section on reward hacking. I wonder what repercussion signal means here, I don't see it mentioned anyw…
↗Not briefed
bluesky
bluesky-ai
2moSirens should sound across the AI industry: 'Anthropic Says It Blocked Possible Efforts to Build Biological Weapons' www.nytimes.com/2026/0…Sirens should sound across the AI industry: 'Anthropic Says It Blocked Possible Efforts to Build Biological Weapons' www.nytimes.com/2…
↗Not briefed
bluesky
bluesky-ai
2mothis feeling is based on how anthropic wants to make "the nicest littlest guy — he's a real listener" and openai wants to make "you seen th…this feeling is based on how anthropic wants to make "the nicest littlest guy — he's a real listener" and openai wants to make "you se…
↗Not briefed
bluesky
bluesky-ai
2moaliás, vamos falar duma coisa importante: a openAI DIZ QUE RESOLVEU. SUPOSTAMENTE resolveu. enquanto nenhum grupo de matemáticos revisar as…aliás, vamos falar duma coisa importante: a openAI DIZ QUE RESOLVEU. SUPOSTAMENTE resolveu. enquanto nenhum grupo de matemáticos revis…
↗Not briefed
bluesky
bluesky-ai
2moTelling an HR recruiter that I literally watched an LLM lie about my resume was like… a great move to build trust imo.Telling an HR recruiter that I literally watched an LLM lie about my resume was like… a great move to build trust imo.
↗Not briefed
bluesky
bluesky-ai
2moPreviously AI-driven research on biological warfare was identified as one of five principal scenarios in which AI could lead to the extinct…Previously AI-driven research on biological warfare was identified as one of five principal scenarios in which AI could lead to the ex…
↗Not briefed
bluesky
bluesky-ai
2moAny industries where you can get away with a false positive, Anthropic is thereAny industries where you can get away with a false positive, Anthropic is there
↗Not briefed
bluesky
bluesky-ai
2moThis is what I have trouble imagining. Like, what can an LLM or even agentic AI actually *do* to destroy the world? Hack the electrical gri…This is what I have trouble imagining. Like, what can an LLM or even agentic AI actually *do* to destroy the world? Hack the electrica…
↗Not briefed
bluesky
bluesky-ai
2moI mean, the rationalists are a cult, and cults do shit like this all the time. Anthropic is full of people who've read Less Wrong, inhaled …I mean, the rationalists are a cult, and cults do shit like this all the time. Anthropic is full of people who've read Less Wrong, inh…
↗Not briefed
bluesky
bluesky-ai
2moSam can't even contain his own AI agents, why would anyone trust him or OpenAI t0o protect the electricity grid?Sam can't even contain his own AI agents, why would anyone trust him or OpenAI t0o protect the electricity grid?
↗Not briefed
bluesky
bluesky-ai
2moAnthropic report on misuse among users — cyberattacks, influence operations, surveillance, bio weapons www.anthropic.com/threat-intel...Anthropic report on misuse among users — cyberattacks, influence operations, surveillance, bio weapons www.anthropic.com/threat-inte…
↗Not briefed
bluesky
bluesky-ai
2moJob #2 would pay more, fully remote unless I want to go into the office, and the interviewer was relieved by my LLM skepticism… but I would…Job #2 would pay more, fully remote unless I want to go into the office, and the interviewer was relieved by my LLM skepticism… but I…
↗Not briefed
bluesky
bluesky-ai
2moJob #1 is not in the tech industry (and would not be paying me enough) and I might not love it long term… and it’s 100% in-office. But that…Job #1 is not in the tech industry (and would not be paying me enough) and I might not love it long term… and it’s 100% in-office. But…
↗Not briefed
bluesky
bluesky-ai
2moHere’s @motherjones.com running a giant Ad for Anthropic disguised as scifi horror porn.Here’s @motherjones.com running a giant Ad for Anthropic disguised as scifi horror porn.
↗Not briefed
bluesky
bluesky-ai
2mothey are right to be scared about AI destroying society. it just won’t be through going SkyNet—it will be through scumbags weaponizing it,…they are right to be scared about AI destroying society. it just won’t be through going SkyNet—it will be through scumbags weaponizin…
↗Not briefed
hn
hn-ai
2moOpenAI pausing new $200 plan subscriptions—
↗Not briefed
bluesky
bluesky-ai
2moGovernment-linked accounts tried to use Claude for work that could lead to bioweapons, Anthropic says www.nbcnews.com/tech/tech-ne...Government-linked accounts tried to use Claude for work that could lead to bioweapons, Anthropic says www.nbcnews.com/tech/tech-ne...
↗Not briefed
bluesky
bluesky-ai
2mothe real question is, would you rather a vibe-coded LLM slop native UI weather app, or a beautiful, handcrafted Electron 1gb RAM usage weat…the real question is, would you rather a vibe-coded LLM slop native UI weather app, or a beautiful, handcrafted Electron 1gb RAM usage…
↗Not briefed
bluesky
bluesky-ai
2mo… and see how long it takes to decide a formally encoded statement of Navier-Stokes: if they're really better than humans at problem solvin…… and see how long it takes to decide a formally encoded statement of Navier-Stokes: if they're really better than humans at problem s…
↗Not briefed
bluesky
bluesky-ai
2moYes, I agree that the refusal to believe how good AI models are at math isn't healthy, and people who are in denial need to move to the “an…Yes, I agree that the refusal to believe how good AI models are at math isn't healthy, and people who are in denial need to move to th…
↗Not briefed
bluesky
bluesky-ai
2moalso they're basically in a cult by now, especially at anthropicalso they're basically in a cult by now, especially at anthropic
↗Not briefed
bluesky
bluesky-ai
2moopenai more like... bad they should call it badopenai more like... bad they should call it bad
↗Not briefed
bluesky
bluesky-ai
2moeu tenho xp científica com IA (sic) e garanto que não é sobre "progresso" mais. Anthropic e outras empresas do ramo sabem que LLM vai dar u…eu tenho xp científica com IA (sic) e garanto que não é sobre "progresso" mais. Anthropic e outras empresas do ramo sabem que LLM vai…
↗Not briefed
bluesky
bluesky-ai
2moyour LLM model is going to make all humans extinct *how*your LLM model is going to make all humans extinct *how*
↗Not briefed
hn
hn-ai
2moShow HN: Nightshift – Your code gets better while you sleep—
↗Not briefed
hn
hn-ai
2moThe Worst Are Full of Passionate Intensity: An Anthropology of AI Safety—
↗Not briefed
bluesky
bluesky-ai
2molike do people not know that nuclear systems are airgapped from the public internet, what's the threat model? someone plugs in a device run…like do people not know that nuclear systems are airgapped from the public internet, what's the threat model? someone plugs in a devic…
↗Not briefed
bluesky
bluesky-ai
2mosounds like anthropic (company) would be better named misanthropic, what with their shit. that they do.sounds like anthropic (company) would be better named misanthropic, what with their shit. that they do.
↗Not briefed
bluesky
bluesky-ai
2moUn investigador de Anthropic dimite y alerta del peligro de la IA: “Puede matarnos antes de 2030” El jefe de Jacob Coxon, joven investigad…Un investigador de Anthropic dimite y alerta del peligro de la IA: “Puede matarnos antes de 2030” El jefe de Jacob Coxon, joven inves…
↗Not briefed
hn
hn-ai
2moShow HN: Claude Code hooks that log every tool call, 124ms per call—
↗Not briefed
bluesky
bluesky-ai
2mocrazy — cognition/windsurf are officially in the game (also, why no xhigh effort level? that’s sick…)🔧 Windsurfcrazy — cognition/windsurf are officially in the game (also, why no xhigh effort level? that’s sick…)
↗Not briefed
bluesky
bluesky-ai
2moMe: "Well, if we treat Knowledge Economy workers as finance, prof services, and universities, then a 20% contraction in about 15% of jobs i…Me: "Well, if we treat Knowledge Economy workers as finance, prof services, and universities, then a 20% contraction in about 15% of j…
↗Not briefed
bluesky
bluesky-ai
2mo2/2 And they’re using those tools to attempt attacks or design weapons that previously lay beyond their capabilities. www.politico.com/ne…2/2 And they’re using those tools to attempt attacks or design weapons that previously lay beyond their capabilities. www.politico.c…
↗Not briefed
bluesky
bluesky-ai
2moLet me tell you a story about how good and cheap Chinese and open weight models are 1. The future of AI subscriptions is 💀 2. They're rea…Let me tell you a story about how good and cheap Chinese and open weight models are 1. The future of AI subscriptions is 💀 2. They'r…
↗Not briefed
bluesky
bluesky-ai
2mo🇨🇳🇷🇺 Anthropic said on Thursday that criminal hacking gangs, Chinese security bureaus, Russia-linked spies and Yemen-based arms manufac…🇨🇳🇷🇺 Anthropic said on Thursday that criminal hacking gangs, Chinese security bureaus, Russia-linked spies and Yemen-based arms ma…
↗Not briefed
bluesky
bluesky-ai
2moI’m not buying this without an independent report or analysis. I’ll update my thinking based on new evidence but “Anthropic said people u…I’m not buying this without an independent report or analysis. I’ll update my thinking based on new evidence but “Anthropic said peo…
↗Not briefed
bluesky
bluesky-ai
2moVery careful parsing. "We didn't use your data *directly*, is not the same as "we used none of your data". qz.com/openai-mathe...Very careful parsing. "We didn't use your data *directly*, is not the same as "we used none of your data". qz.com/openai-mathe...
↗Not briefed
hn
hn-ai
2moA somewhat optimistic view of AI in mathematics—
↗Not briefed
bluesky
bluesky-ai
2moyou can feel this happening when it happens and it sucks, all of a sudden your LLM is hyperfixated on something trivial or probably wrongyou can feel this happening when it happens and it sucks, all of a sudden your LLM is hyperfixated on something trivial or probably wr…
↗Not briefed
bluesky
bluesky-ai
2moIl va falloir re-regarder cet épisode de série quand OpenAI aura publié une preuve de la conjecture de Hodge en piquant plein d'idées au pa…Il va falloir re-regarder cet épisode de série quand OpenAI aura publié une preuve de la conjecture de Hodge en piquant plein d'idées…
↗Not briefed
hn
hn-ai
2moShow HN: Pixel Agents – A pixel-art mission control for your Claude Code agents—
↗Not briefed
bluesky
bluesky-ai
2moMy main point though is that I think that this argument is mostly about a refusal to believe the models are as good as they are. I’m very h…My main point though is that I think that this argument is mostly about a refusal to believe the models are as good as they are. I’m v…
↗Not briefed
hn
hn-ai
2moShow HN: iTerm2 Plugin for Codex/Claude CodeI made this beta iTerm2 plugin with Codex. It adds Codex sessions to the Session Status. The sidebar can also show Claude Code session…
↗Not briefed
bluesky
bluesky-ai
2moTraining an LLM on nothing but Garak from DS9Training an LLM on nothing but Garak from DS9
↗Not briefed