Square
Following
Hot
News
Notifications
Profile

Bitduke

vip
No content yet
Peak Tier VIP 27.3 Year(s)
0
Following
16
Followers
108
Liked
we're getting dangerously close to simulating entire cities with AI Fable 5.2 recreated Tokyo's Shibuya in just 12 hours, complete with NPCs that have simulated behavior imagine what these simulations will look like in another year
the best maze runner is still Fable 5.2, and apparently still unbeaten on MazeBench the two fresh models, GPT-6.2 Astra and Gemini 4 Neon, didn't manage to catch up on this benchmark of long-term planning, spatial reasoning and increasingly difficult 3D puzzles. although Gemini
Google just shot up the RSI Index benchmark with Neon, getting closer to Anthropic (at a fraction of the cost) Neon reached #4 on the benchmark, above Astra 6.2 and Google's previous best model, Argon (with most of its gains coming from post-training). really good results, but
GOOGL-3.48%
turns out the rumors about Anthropic canceling its compute deal with SpaceX are probably false, as Musk himself said he thinks that’s unlikely and Anthropic actually wants to rent even more compute from them according to him, Anthropic is looking to secure an additional 10 GW of
ANT-13.13%
SPCX-3.69%
Anthropic shares are expected to open at $100 on Nasdaq, up from their $83 IPO price (~20.5% increase)
ANT-13.13%
Gemini 4 Sodium is coming and 114 more models are on the way Google is really cooking something and releasing models at a much faster pace than before, while quality is also improving (Neon is quite good)
GOOGL-3.48%
NEON-5.64%
top 10 tasks where agents struggle and top 10 where they succeed OpenAI just released a leaderboard of research tasks agents complete with zero human help, using Epoch AI’s version of O*NET (the US labor dept’s way of breaking jobs into tasks) the spread is quite visible,
the most interesting line in OpenAI’s new report: human time on safety & alignment grew 77% from Jan to Sept, the biggest absolute increase of any research area part of it is new hires ofc and part is existing researchers moving over, and the growth picked up after the Hugging
the state of managing ai agent fleets inside OpenAI (these fleets just keep getting bigger) their latest research acceleration report, back in May most OpenAI researchers were running fewer than 4 agents at once (!) but by mid-october over 2/3 hit a daily peak of 4 or more, and
Genie 4 (available to Google AI Ultra subs) can generate a deer in real time with just one prompt while you're exploring the generated world
GPT Astra and Sol 6.2 just scored lower on SimpleBench than the models they replaced, roughly 5 to 6 points down (Fable 5.2 is still the king) SimpleBench tests everyday spatial, temporal and social common sense, with deliberately tricky wording. it’s the kind of stuff a regular
another interesting finding from the recent OpenAI research report: the median OpenAI researcher now uses $2,400 worth of ai agents PER DAY, and for the top researchers, that number reaches $23,000 (!) based on API list prices tho, but still pretty crazy considering both numbers
Gemini 4 Neon x Antigravity created a black hole
so even the estimated doubling time is increasing now, from 2.2 to 3.8 months (progress is still exponential, just a bit slower than previously estimated)
ai agents can now finish almost HALF of OpenAI's 2–4 hour research tasks without human intervention, up from just 23% in march. that's roughly a 2x improvement in only 7 months (!) OpenAI just published a new report on how AI agents are accelerating its own research and
interesting take on why reviews of the new Astra are so mixed
right, don’t take yourself out of the thinking process
no H200 (or Hopper-generation chips) for China? around 600,000 H200 chips are reportedly stuck in Hong Kong, awaiting permission to enter mainland China
Load More