#open-ai — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #open-ai, aggregated by home.social.
-
watching “Hilarious Nuclear Criticality Incident at Los Alamos” with Pu rods; Right at the end, the podcast author talks about his industry, the north sea Oil industry and how if he’d stayed as an Engineer he’d spend the rest of his career doing cleanup, decontamination and disposal instead of the *sexy* Engineering dreamed up in University.
#Criticality / #SystemsSafety / #LosAlamos / #Sellafield / #NeutronReflection / #IncidentResponses / #RootCauses / #AlexanderTheOk <https://www.youtube.com/watch?v=SVoo3tcYdr8>
After watching the #OpenAI, #Anthropic, #NVidea shitstorm this year, a lot of *AI engineers* will also meet this fate. Cleaning up the toxic AI mess left behind.
-
Lawmakers blast AI companies after researcher warns of human extinction by 2030
By Dara KerrFormer Anthropic employee Jacob Coxon said AI will become ‘superhuman systems’ that can cause human extinction by the end of the decade
https://www.theguardian.com/technology/2026/sep/09/lawmakers-blast-ai-human-extinct-2030
#AIartificialintelligence #Technology #Anthropic #USpolitics #USnews #OpenAI #BernieSanders #TedCruz #ElonMusk #TheGuardian #DaraKerr
-
Lawmakers blast AI companies after researcher warns of human extinction by 2030 https://www.theguardian.com/technology/2026/sep/09/lawmakers-blast-ai-human-extinct-2030 #AiArtificialIntelligence #Technology #Anthropic #UsPolitics #UsNews #Openai #BernieSanders #TedCruz #ElonMusk
-
OpenAI released ChatGPT Images 2.5 on September 8, 2026, claiming up to 50% faster image generation. A new Sketch tool lets users draw references directly in ChatGPT. The move follows Microsoft's shift toward in-house image models. https://www.implicator.ai/openai-ships-chatgpt-images-2-5-with-sketch-tool-and-up-to-50-faster-generation/ #AI #ImageGeneration #OpenAI
-
Za Christopher Keruac:
"OpenAI ogłosiło, że ich agenci AI rozwiązali Problem Milenijny Naviera-Stokesa.
Świat zachwyca się AGI. W rzeczywistości za kulisami rozegrał się thriller: kradzież badań naukowców, korporacyjny szantaż i groźby niszczenia karier.To nie AI wymyśliło ten dowód.
Metodę stworzyli Diego Córdoba i Luis Martínez-Zoroa. Matematycy Tristan Buckmaster i Levent Alpöge rozwijali ją przez rok, używając AI jako narzędzia i weryfikując wyniki w Lean. 15 sierpnia osiągnęli przełom.
Buckmaster płacił za OpenAI Codex z własnych grantów. Wrzucał tam wszystkie szkice i notatki z całego roku.
Gdy na początku września plotka o ich przełomie dotarła do OpenAI, w firmie wybuchła panika. Rzucili gigantyczny compute, by pójść dokładnie tą samą ścieżką.
6 września do Buckmastera dzwoni Sébastien Bubeck z OpenAI. Twierdzi, że model sam wpadł na rozwiązanie.
Gdy Tristan pyta wprost: „Czy trenowaliście model na moich prywatnych sesjach w Codexie?”, po stronie OpenAI zapada absolutna cisza.
Zaczyna się presja. Bubeck żąda wykreślenia Leventa z autorstwa, bo pracuje w Anthropic i psuje PR OpenAI.
Proponuje układ: poparcie dla nich do Nagrody Claya w zamian za solowy sukces AI. Tristan odmawia.
Bubeck: „Dlaczego rujnujesz swoją karierę? Nie muszę być miły”.
Bubeck próbuje rozbić duet, pisząc do Leventa, że „Tristan nie myśli racjonalnie”. Bez skutku.
Mimo braku zgody OpenAI publikuje wpis na X: „Nasz model i agenci sami rozwiązali problem!”, marginalizując wkład autorów metody: Córdoby i Martíneza-Zoroa.
OpenAI zaprzecza podglądaniu sesji, ale w oświadczeniu kluczy: „Nie możemy wykluczyć, że zanonimizowane dane z użycia produktów pomogły modelom”.
Pytany wprost przez Buckmastera o to, czy model był trenowany na jego notatkach z Codexa: Bubeck nie zaprzeczył.Jedyna nauczka z tej dramy:
Gdy wrzucasz nieopublikowane idee do zamkniętego AI, stajesz się darmowym dawcą danych. To nie był triumf AGI. W wyścigu po wycenę przed IPO korporacja bez skrupułów przerobi Twoje życiowe odkrycie na marketingowy slajd." -
Dear Scam Altman. I asked AI about LLMs and AGI. Turns out you lied. Again. 😆 (transcript in alt text x 2) #AI #openai #samaltman
-
AI Agents Are Not Going “Rogue,” But Recent Developments Lead Increasingly-Concerned Researchers To Call For AI Slow Down
-
OpenAI's announcement Tuesday that it has solved one of mathematics' legendary Millennium Prize problems should have been a moment of triumph. The result is both an undeniable achievement and a striking demonstration of just how rapidly AI is transforming mathematics. But before it was even formally announced, the breakthrough had been complicated by the unusual […]#AI #OpenAI
OpenAI’s sly mathematical breakthrough sends a chill through academia -
Imagine if there was affordable (or even free) health care instead...
Man told ChatGPT he was feeling delusional. ChatGPT insisted he was Jesus.
-
#jacobcoxon quit #anthropic and #openai over safety concerns that it could kill us all in a decade. Gary Marcus NYU professor calling for a #consumerboycott Hmmm, we can’t get people to boycott the #propagandist #Twitter. You think they’re going boycott AI?
-
OpenAI recently announced that its model has solved one of the most difficult mathematical problems in the world. So who gets the credit for the breakthrough?
https://www.plagiarismtoday.com/2026/09/09/who-gets-credit-when-ai-makes-a-breakthrough/
-
There is only a potential 10% chance that #AI will destroy humanity within a decade‽ As in 2036‽ Good to know.
Jacob Coxon (former #OpenAI / #ChatGPT & #Anthropic / #Claude employee) claims both companies are are either willfully ignorant of the dangers of super intelligence or believe they can magically control the crazy beast by helping it achieve hyper intelligence.
More alarming is the response by a current Anthropic employee.
Link: https://x.com/hilbertspaess/status/2097476196791709843
-
Patrick G. Eddington
@pgeddington.bsky.socialThis particular paragraph re: AI superintelligence actualization really does have a Cyberdyne Systems quality to it.
https://paulfchristiano.substack.com/p/personal-statement-on-joining-the
-
...'OpenAI' scheint uns zum Narren zu halten und #srf scheint Fan von 'OpenAI' zu sein
-
Sure, we solved the problem, but probably ripped off someone else's private work in the process. But who cares?
VentureBeat: OpenAI solves longstanding math problem with 10,000-agent swarm — but can't rule out benefiting from a researcher's private Codex data https://venturebeat.com/technology/openai-solves-longstanding-math-problem-with-10-000-agent-swarm-but-cant-rule-out-benefitting-from-a-researchers-private-codex-data @venturebeat #OpenAI #privacy
-
Anthropic Researcher Quits, Warning the AI Race Could End in Human Extinction
#AI #anthropic #OpenAI #artificialintelligence #TechNews #agi
https://firethering.com/anthropic-researcher-ai-race-extinction/
-
UK Lawmakers Are Freaking Out Over AI’s Summer of Chaos
-
Il professore di matematica della New York University Tristan #Buckmaster ha annunciato martedì tre dimostrazioni con una scoperta preliminare su uno dei principali problemi irrisolti della matematica teorica. I risultati, realizzati in collaborazione con il matematico di #Anthropic Levent #Alpöge e utilizzando modelli di intelligenza artificiale Codex e Claude, sono significativi di per sé — ma sono anche accompagnati da un'insolita controversia che circonda i tentativi di OpenAI di risolvere lo stesso problema.
Secondo la dichiarazione, un'iniziativa parallela di #OpenAI si è basata sul loro lavoro prima che diventasse pubblico, dando origine a un groviglio di rivalità accademiche e affermazioni contrastanti.
-
OpenAI says it solved Navier-Stokes, one of the seven Millennium Problems from the Clay Institute.
Specifically, they used an internal model not yet public (more capable than GPT-6 Astra), running up to 10,000 agents in parallel, which produced the proof of existence of a singularity in 88 hours and formalized it in another 17. Estimated compute cost: around $15 million.
The catch: they solved the "forced" version of the equations (fluid with an external force applied in a controlled way), not the "unforced" version, which is the only one the Clay Institute recognizes for the prize. So even if they wanted to, they don't win the million dollars, and sure enough they've said they're passing on it anyway... draw your own conclusions.
On top of that, Tristan Buckmaster (NYU) and Levent Alpöge (who, funnily enough, works at Anthropic) had been on this for a year, working on the Euler equations (close relatives, not identical). Buckmaster is now accusing OpenAI of drawing on his unpublished work through his Codex account, with a math strategy identical to his own. A few hours before OpenAI's announcement, the two had posted three intermediate steps of their solution online.
I'll leave the conclusions open. The only genuinely sensible take is Roberto Natalini's (CNR): the risk is that researchers stop sharing intermediate results online, afraid of being scooped by whoever has 10,000 agents and $15 million in compute.
Which is ironic, since open science is exactly what made training these models possible in the first place.
-
I Let an AI Agent Hack All My Gadgets—and I’d Do It Again
https://fed.brid.gy/r/https://www.wired.com/story/i-used-ai-to-hack-my-home-network/
-
"Stanno giocando d'azzardo con le nostre vite". Sono queste le parole di Jacob Coxon, che dopo tre anni di lavoro come ricercatore in Anthropic e OpenAi, ha scelto di rassegnare le proprie dimissioni. Con un post su X ha spiegato i motivi della propria scelta. I toni sono inquietanti. "Oggi mi sono dimesso da Anthropic - ha esordito -. Ho trascorso gli ultimi tre anni a fare ricerche sul pretraining sia a OpenAI che ad Anthropic. Nessuna delle due aziende sta agendo in modo responsabile. Stanno correndo dritte verso una super-intelligenza automigliorante".
1/3
#Anthropic #OpenAI #ChatGPT #Gemini #Claude #JacobCoxon #AI #IA
-
When a child starts trying all possible answers to a test question, we go "you've clearly not understood the fundamentals, let's go over them."
When a fucking LLM "agent" does the same thing, not only people let it go until it stumbles upon the correct answer by chance, they start forming a religion around it.
Astonishing, really.
-
🤖 【OpenAI News】Paul Christiano joins OpenAI Foundation Board
Paul Christiano joins the OpenAI Foundation Board and its Safety and Security Committee, bringing experience in AI align...
🔗 https://openai.com/index/paul-christiano-joins-openai-foundation-board
-
400 Seiten geheime Verträge zeigen die enge Kooperation zwischen #Pentagon und #KI-Firmen wie #OpenAI und #Google. KI-Technologien werden für Überwachung, Zielerfassung und autonome Militärstrategien entwickelt. Hersteller beraten das US-Verteidigungsministerium und erhalten Zugang zu sensiblen Daten. https://www.heise.de/news/Pentagon-Leaks-Die-verheimlichten-Vertraege-der-KI-Giganten-11445865.html?seite=all