Haider.
@haider1
open intelligence for a decentralized future
4.2K Following 71.6K Followers
openai devday is in 2 days, and employees are already vagueposting
so far from hints and leaks we've got:
- astra minor
- ultrafast mode
- possible pro max tier
- voice + agent integration
- astra successor and their first hardware demo
- "o" / aeon, an always-on assistant with persistent agents
i'll be over the moon if we get gpt-oss-2
Show more
openai absolutely cooked with the new chatgpt UI
probably the best visual direction they've taken in a while, and you can clearly see them trying to merge chat, work, codex, files, etc. into one coherent product
still needs some work on:
fixing long-chat navigation
keeping the sidebar consistent
making files/projects easier to manage
Show more
openai absolutely cooked with the new chatgpt UI
probably the best visual direction they've taken in a while, and you can clearly see them trying to merge chat, work, codex, files, etc. into one coherent product
still needs some work on:
fixing long-chat navigation
keeping the sidebar consistent
making files/projects easier to manage
Show more
the more i think about Astra’s successor, the crazier it sounds
this model started training on august 28, resolved navier-stokes using a 10,000-agent system, and solved 100+ long-standing open problems across mathematics -- while training is still ongoing
and this model is already old news internally by now
Show more
the more i think about Astra’s successor, the crazier it sounds
this model started training on august 28, resolved navier-stokes using a 10,000-agent system, and solved 100+ long-standing open problems across mathematics -- while training is still ongoing
and this model is already old news internally by now
Show more
gpt-6 sol is massively underhyped
it's somewhat better than gpt-5.6 sol while being cheaper, much faster, and keeping most of that same feel -- maybe it's basically 5.6 with more RL + some astra distillation mixed in
either way, i think we get a gpt-6.1 drop by oct which pushes much closer to Astra with a lot more RL
Show more
elon, this isn't as simple as it looks
6 months from now, openai and anthropic will be approaching or might have achieved RSI, and oai will already be pushing toward GPT-6.5 by then
unless both run into massive compute bottlenecks, i don't see how SpaceXAI closes that gap
Show more
1. We will keep accelerating. Our AI efforts are only 3 years old, vs 6 and 10 years old for Anthropic and OpenAI. If our second derivative remains strong, SpaceX will reach pole position in about 6 months.
2. Once you far exceed the caliber of intelligence needed for a class of tasks, additional intelligence is pointless. You don’t need (and it would be cruel to put) Newton-level intelligence in your toaster!
3. Hardware is hard. Bringing massive compute online rapidly is incredibly difficult. SpaceX has demonstrated exceptional ability in this regard and will only get better.
Show more
behind the scenes from the Trump-Xi meeting:
Xi was reportedly trying to discuss the bigger questions around ASI development in the coming years -- but Trump appears to have interpreted it instead as support for his idea of calling AI "Super Intelligence"
weird obsession with renaming things
Show more
behind the scenes from the Trump-Xi meeting:
Xi was reportedly trying to discuss the bigger questions around ASI development in the coming years -- but Trump appears to have interpreted it instead as support for his idea of calling AI "Super Intelligence"
weird obsession with renaming things
Show more
i personally love GPT-6 Astra, but openai should keep Astra's successor internal for research and use it to distill cheaper models
astra 6.1 is much better than 6.0, so the compute situation gets even uglier -- oai needs the next model to be better than astra, cheaper per task, and maybe 5-10x cheaper on compute internally
nvidia Rubin will help, but that rollout takes months. user demand can explode in weeks
that gap just isn't sustainable
Show more
gemini 4 is probably coming next month at this point
considering how absurdly fast google went from flash 3.6 → 3.7 → 3.8, with each one being a legit huge jump in coding, i think there were only like 3–4 weeks between releases
they must be grinding 24/7 over there.
Show more
gemini 4 is probably coming next month at this point
considering how absurdly fast google went from flash 3.6 → 3.7 → 3.8, with each one being a legit huge jump in coding, i think there were only like 3–4 weeks between releases
they must be grinding 24/7 over there.
Show more
you really don't need frontier models like opus 5.5 and gpt-6 astra for everything
and that's why people are considering going local, especially now that qwen 4 27b is coming -- codex is amazing, and sol 6 is cheaper now too
enterprise is clear, but for $200 plans and regular paid users, it's still a mystery
Show more
you really don't need frontier models like opus 5.5 and gpt-6 astra for everything
and that's why people are considering going local, especially now that qwen 4 27b is coming -- codex is amazing, and sol 6 is cheaper now too
enterprise is clear, but for $200 plans and regular paid users, it's still a mystery
Show more
the upcoming chatgpt pricing structure is basically:
$20 = the new free tier
$100 = the new plus tier ($20)
$200 = the new pro 5x tier ($100)
$500 = the new pro 20x tier ($200)
as simple as that
Sam Altman says:
"the AI industry must not accept too much technological risk just because the benefits feel too important to slow down. beating competitors in a race is not a reason to make rash decisions"
We have unilaterally slowed down in the past, and we will do so again
Show more
openai clearly chose efficient models like gpt-6 sol/luna to free up data center capacity for bigger models
we know they have a larger model than astra that they'll eventually release
but opus 5.5 using 4x more output tokens is eye-opening. if API price roughly reflects model size, that could mean up to 8x more compute at those benchmark scores
Show more
openai clearly chose efficient models like gpt-6 sol/luna to free up data center capacity for bigger models
we know they have a larger model than astra that they'll eventually release
but opus 5.5 using 4x more output tokens is eye-opening. if API price roughly reflects model size, that could mean up to 8x more compute at those benchmark scores
Show more
sapceXAI clearly fumbled with Grok 4.7 after the momentum they had with Grok 4.6
i hope this is a one-off, similar to how anthropic made mistakes with opus 5 before mogging everyone with opus 5.5 -- however, Grok 4.8 will be using a new C++ stack and 2.5T parameters
so it could be something
Show more
openai devday is in 4 days, but i doubt it brings a much better model
gpt-6 astra, along with 6 sol/luna, only just launched. i think oai underestimated anthropic's release pace
i'm expecting mostly tools, and likely a demo of their first hardware and an astra successor model that has already resolved navier-stokes and 100+ math problems
Show more
