Jj McMc

they pay me to post here

@haider1 Damn, an openai research used an internal model?? Crazy news!! You keep tweeting stuff that is very well known or just plain uninteresting all the time, just reiterating blog posts and tweets everyone else has seen anyway

next month looks huge for big model releases: oai researcher is already using an internal model (likely gpt-5.6) as her daily driver elon says the 1.5t v9 grok model has finished training (likely grok 4.5 / 5) gemini 3.5 coming next month anthropic says the "mythos" class model will release soon

@uncreativetom That would make it too easy for their customers

can someone from germany please explain why so many deutsche bahn trains can’t be booked on the deutsche bahn app

@DavidSHolz Colossus 1

how many gpus do you think a single researcher can handle at once for a single big training job without any help? (assume it's set up with slurm or something)

@GaryMarcus @theo Gary, shut the fuck up.

@theo dude, do a little homework? paul graham wrote something brilliant that you should read: paulgraham.com/disagree.html then look at the thing i attacked and reference the pyramid. and LEARN something.

Fuck this OpenAI employee, seriously fuck him. Also read this quantitative study, which I had nothing to do with, that says my technical predictions have bee largely correct. github.com/davegoldblatt/… I should not have to put up with this level of unhinged hostility, simply for defending views that have frequently been challenging to his company. Nor should I have to put up with him neither giving evidence nor allowing me to respond. This man appears to be both a coward and a liar, with only insults, literally a bottom feeder in @Paulg’s pyramid of argumentation. That this empty slander is all he has got is perhaps a sign of how desperate OpenAI has become.

@evandotsh @FioraStarlight Oh i dont want to use them, im just interested in their architecture. Its weird not knowing basic facts like parameter count or even if a model is an MoE for models that were as groundbreaking as o1

@AccountMus629 @FioraStarlight I hear this a lot, and my question is why? We already have open source models that are basically as intelligent as the ones we had 6 months ago. Why would we need year old models to be open sourced?

@OfficialLoganK @roboflow #Logangforlife #LoganPaul

@OfficialLoganK @roboflow Damn, you certainly found a benchmark without a "cost" column, good job logang

Gemini 3.5 Flash outperforms 3.1 Pro on many vision use cases (like the below Roboflow eval) while being ~6x faster on average 🤯 Gemini multimodal understanding for the win.

@FioraStarlight It feels so weird to know basically nothing about models we used ages ago

@FioraStarlight Them and openai should at some point open source them i think. I dont care about 4o or anything like that, but it just feels weird that models will just be lost to time and even the knowledge about them - e.g. id love to know at some point how many params o1 or haiku models are

@scaling01 Tbf they are probably just doing it to get more training data

DeepSeek is the only lab that is still trying to make intelligence too cheap to meter

@AlexanderKalian @SprengmEAcc You are asking pointless and ridiculous questions - "whats the scaffolding" is like asking if a human who solved it has a calculator on his desk or used python - it doesnt matter what tools they used, this is maths, it is objective, the only thing that matters is the proof

@AlexanderKalian @SprengmEAcc My god you are fucking stupid RAG - There was no RAG, very obviously. "Prior prompt engineering" - Again, the prompt is in the paper, it is simply the problem they want it to solve. Nothing more. The CoT isnt public because it sounds very different to human speech in 5.5

When OpenAI claimed their "internal model" autonomously one-shotted an Erdős problem, they were really making two big claims: 1) A frontier LLM solved an Erdős problem. 2) It did so fully autonomously, without significant human guidance. Claim 1 has decent evidence - the solution was verified by world-renowned mathematicians. Claim 2 has almost none. We have only high-level anecdotal tweets from OpenAI executives. "Source: trust me bro." Claim 2 is by far the more important one. If true, it would suggest LLMs can now generate truly novel, out-of-distribution research breakthroughs - with major implications for human researcher automation, recursive AI self-improvement, and OpenAI's valuation ahead of IPO. If the model was guided by humans (via prompting, scaffolding, chain-of-thought, or hidden training data), then it's still an impressive leap - but far more incremental and expected. Until OpenAI releases the model or publishes a detailed, peer-reviewed study, Occam's razor suggests the less extraordinary explanation. Especially from a company with both a track record and strong financial incentive to overhype. Extraordinary claims require extraordinary evidence.

@rowanfornow This question is also stupid af, so i just get wet for 30secs every day for 5 years? There is no was to drown in 30 secs and nothing is going to eat you, there arent that many predators that eat humans in the sea

genuinely, ChatGPT might have the most infuriating writing pattern I've ever seen. why does it do this. who talks like this.

@brown_john18944 @iv_sal1 this is just such a stupid take, ALL games were programmed ON pc and work there IN THE FIRST PLACE

@iv_sal1 meanwhile the actual pc experience

Man, i wish my PC was plug and play. I cant believe i need to do all of this everytime I want to play a game.

@iruletheworldmo Soooo?

happy 5.6 day everyone, now we can all solve math.

@StrasserGregor @FlorianGallwitz @Cire_Legan Cooler Zirkelschluss

@FlorianGallwitz @Cire_Legan Auf welcher Grundlage soll ein LLM einen Vorgang bewältigen der Intelligenz voraussetzt, welche es nicht haben kann? Die Frage führt nur in den Irrgarten.

War Einstein wirklich intelligent? Hat er wirklich neue Ideen generiert, oder hat er nicht einfach nur Vorhandenes neu gemixt?

@StrasserGregor @FlorianGallwitz @Cire_Legan Was ist das denn für ein dummes Gelaber, nur weil man sagt dass Silizium auch intelligent sein kann heißt das nicht direkt dass man was gegen Menschen hat?

@FlorianGallwitz @Cire_Legan Eine Masse Silikon kann nie in der Form intelligent sein wie ein Mensch. Wenn sie das ignorieren, was ist dann Ihr Menschenbild?

@Cire_Legan @FlorianGallwitz Hat sich schon geändert, und auch nicht erst heute

@FlorianGallwitz 👉>>Kann sich aber ändern<<

@gerardsans @NumerusOpinions @GaryMarcus You are also moving the goal posts soooo much, who gives a shit if it is just sophisticated brute force (it isnt), the result is still that we are able to prove stuff we havent so far, and capabilities are only improving. Who cares if it is "proper" intelligence, it gets results

@gerardsans @NumerusOpinions @GaryMarcus Lol you dont have a clue what you are talking about First up the prompt was simply just the problem they wanted it to solve. No prompt engineering. Secondly, yes, normal people can also almost get there using chatgpt. Many math problems slightly have been solved with it

is the new math result neurosymbolic with Lean, harnesses etc or a pure LLM?