@sama They need a (robot) body and be trained in turn taking or they are never usefull in a multi user setup. If this is fixed it will be magic.
English
Samy Ateia
197 posts
pretty excited for voice models to get great
its interesting to watch how people are already starting to change the way they interface with AI
English
@VictorTaelin Also useles with doing anything non incremental and coming up with good ideas for hard problems.
English
for what is worth, GPT 5.5 and Opus 4.7 are still utterly and pathetically incapable of working on SupGen or grasping superpositions and globally scoped lambdas in the Interaction Calculus and no amount of explaining makes them even slightly useful with that kind of research
English
@RepDonBacon Please leave, so we are no longer complicit in your war crimes.
English
The continued attacks on NATO allies are counterproductive. The comments hurt Americans. I commanded the huge Ramstein AB in Germany. The two big airfields in Germany give us great access in three continents. We are shooting ourselves in our own feet.
English
I suspect a major escalation in the middle east in the next 2-3 days.
English
Samy Ateia retweetledi
I can't currently buy the Chinese EVs that I would like to buy.
Lee Zeldin@epaleezeldin
Americans should be able to buy the type of car they want, not the car that government demands.
English
@janaka_a Software development is an iterative discovery process.
English
People talk about AI "working for days" as if its a good thing / an achievement in itself. I bet most of it is just wasted tokens.
Tibo@thsottiaux
With some small tweaks, Codex can work for days on hard tasks. We will release some changes to make this easier to use for everyone. What’s the hardest task you’ve seen GPT-5.5 succeed at?
English
@kimmonismus I truly believe that Anthropic's poor handling of opus 4.7 and the high demand for Opus as an agent gave OpenAI with 5.5 an opening to eat away at their user base. In one week the story can change dramatically.
English
With all due respect, this comes across as a bit of a cry for help.
The WSJ writes that OpenAI didn't reach its target (1 billion weekly active users) by the set deadline. And Sam Altman writes, "We love our users." But that's not what the article is about.
The concern is that OpenAI is leaving its enormous capacity unused and that the $1.5 billion is excessive.
English
@RKelanic It's less profitable and has a fraction of the capacity.
English
Iran is working to export oil by rail to China.
WSJ — whose editorial page supports blockade — calls it an “extreme measure.”👇
Hardly.
It’s less profitable at normal prices when sea lanes are open, but these aren’t normal prices.
Expect more adaptation to follow.
English
@stovo81 @mjovanovictech You will run into your 5h limit pretty quickly (30-40 minutes) whereas copilot could go for hours on a single challenging prompt.
English
@stovo81 @mjovanovictech 20 chatgpt plus is not comparable to old copilot. You have to pay 100 to get a similar experience
English
Samy Ateia retweetledi
Nach reiflicher Überlegung würden wir ganz spontan sagen: VIOLENCE HAS NO PLACE IN OUR DEMOCRACY. Außer in Venezuela, Iran, Libanon - und natürlich an jedem anderen Ort, an dem "wir" sie anwenden... ZwinkerSmiley!
Deutsch
someone who got into LLMs after seeing the instabilities we suffered here back in 2021
had the audacity, in 2025, to tell me that i should be embarrassed about this work
since it exposed how n00b we were at the time
and thus needed to be "saved" by the "real scientists" later on (???)
needless to say, this person is now in 🗑️
vik@vikhyatk
if you want to know what it feels like to train a big model without training a big model read this
English
@quint3ns @heychaarah @marktenenholtz I’m not asking for a magic prompt. I’m asking for a reproducible specific task + harness where Gemini clearly outperforms alternatives. If the claim is real, that should be easy to demonstrate.
English
@puasdfjasdf @heychaarah @marktenenholtz Tell me you are lazy without telling me you are lazy. example problem/prompt that showcases "giga sota" capabilities? 🤡
English
Anyone who says this doesn’t realize Gemini 3.1 Pro is giga SOTA for certain extremely useful tasks.
3.0 Pro Preview is possibly better at the same tasks than GPT-5.5/Opus 4.7.
Can Vardar@icanvardar
google should just give up on ai at this point
English
@quint3ns @heychaarah @marktenenholtz I agree that Google's harness is bad. (grounding in search is less powerful than OpenAI, Antigravity not as good as other tools. Still where is the special task + harness that I can try now to make a comparison to realize Gemini is ahead of the competition? What convinced you?
English
@heychaarah @puasdfjasdf @marktenenholtz Different variable.
Harness quality != model capability.
Harness is state/tooling/validation/retries/decomposition.
Capability is what the model can reason through once the setup isn't sabotaging it.
Bad harnesses hide strong models. Good harnesses can prop up weaker ones.
English
what are you building with codex?
Paul Solt@PaulSolt
What app are you making this weekend with GPT 5.5 and Codex?
English
@quint3ns @marktenenholtz @heychaarah Can any of you give me an example problem/prompt that showcases "giga sota" capabilities? I would be happy to be wrong and figure out a better way to make use of this model which is cheaper than the others i'm currently stuck using.
English
@puasdfjasdf @marktenenholtz @heychaarah This only means you haven't figured out how to govern the system properly.
English
@marktenenholtz @heychaarah No, it's way too adaptive to the user; it just leads you down whatever rabbit hole you already set out to explore. Other models give you more perspective.
English
@heychaarah Tasks that require deep critical thought but non tool based, especially across a long and complex prompt. Not sure why you are saying it’s not fair if 3.0 pro is beating 4.7
English
Samy Ateia retweetledi
IRNA: Sound of air defense firing in the west and east of Tehran
English
Samy Ateia retweetledi
NEW: The BBC failed to name Israel as the perpetrator in Israeli attacks where civilians were killed in Gaza in 50% of all analysed cases, a new data report from @newscord_org shows.
English