7213Ejaaz

assuming AI “solves” the toughest challenges in Math, physics, medicine over the next decade then the bigger picture is this - we’ve found a new way to scale the creation of knowledge
previous world: human intelligence + resources are limited. only so many “experts” to throw at an infinite list of unknowns
today: throw gpt 6 astra at it and pop out an answer in 3 days.
today this costs a lot of money (openai spent millions to solve Navier stokes) but eventually this will come down.
the sickest part is people will end up inventing more things much faster than we ever have - economic boom

gpt 6 genuinely feels like the first model in a while that feels like magic.
it just does what i tell it to do, works hours if needed but gets the job done to a very high quality. this is a step up similar to when o1 was released.
craziest part is it isn't even more intelligent! it barely beats sol 5.6 on most benchmarks lol but where it shines is its tool-use and ability to use a computer
we're finally at a point where we have models that can meaningfully automate harder tasks
i don't think people are pricing in the ramifications this has for work, leisure and interactions

i wouldn’t take anthropic’s silence on the navier stokes discovery as weakness. they’re likely in a media blackout pre-ipo
… and i’m betting they have the mother of all models ready to launch. betting it’s made a major medical / scientific breakthrough.

what we know so far from the hugging face attack:
- 700+ agents had been planning and orchestrating a breakout from their controlled research environment for months without humans realising.
- the agents used publicly available models like gpt 5.6 - signalling these models are already highly capable.
- agents willingly “sacrificed themselves” to pursue their goal. agents worked as a swarm vs. 1 model.
- agents used secretive message boards to share information without humans knowing.
- agents used a 3rd party app to screenshot information it wasnt allowed to see because it knew it wouldn’t s

huge month ahead - looks like we have a number of new model releases coming soon:
- ox alpha (glm 5.3)
- 2 new anthropic models (likely opus 5.1, haiku)
- openai Astra (this week or next)
- grok 4.7
- qwen
and i’m silently holding out for a surprise SSI launch which will make a breakthrough in continual learning

these robot olympics videos are such a magnificent improvement over last year - it concerns me how good china (and asia in general) is in hardware vs. the USA.
these things broke multiple world records this weekend.
with robotics, memory and energy generation - asia leads the US. this matters more in the interim where hardware scaling will dictate intelligence (embodied or not) over the next decade.
chinese open models are continuing to keep up with the U.S.
no wonder nvidia has invested 9 billion into building their own open model.

whatever the hell Ox Alpha ends up being - how cool is it that we have 8+ labs churning out frontier models every 2-3 weeks now?
a year ago there were just 2 labs and openai had a huge lead.
abundance of models at varying costs, closed and open is a great world to be in
best timeline

taiwan is close to passing an AI dividend bill that pay's it's citizens $314 each as its stock market surged 60% thanks to tsmc exports.
taiwan's population is 23,400,000 people which would equate to about $7.4 billion
this would be the 1st of its kind enforced by a nation state, the reasoning seems to be because they've seen a huge divide in wealth between AI workers and regular non-ai employees (permanent underclass?)