Square
Following
Hot
News
Notifications
Profile

Bitduke

vip
No content yet
Peak Tier VIP 27.3 Year(s)
0
Following
16
Followers
106
Liked
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
GPT-6.2 Astra can solve math problems nobody had solved before, yet still struggles with relatively simple research tasks pretty interesting results from Epoch AI's independent evaluation, Astra scored 174 on the ECI, tying Fable 5.2, but fell behind on realistic research tasks
recently, i've started seeing more and more real-world benchmarks (like this one) that test frontier models on actual engineering tasks in physical labs honestly, i'm not surprised. as automated r&d keeps getting better, we're moving closer to the next phase, where ai models
google definitely has every chance to compete in the ai race
GOOGL+0.76%
Astra 6.2 also did pretty well on banteg’s sokoban benchmark (sokoban is NP-hard, and one bad push can trap you later), clearing all 66 rooms about 33 minutes after fable 5.2 for about $20 to Fable’s $52 also, Fable 5.2 finished in 198 turns and Astra took 540 here. with all
in 85% of runs, GPT-6.2 Astra wondered if it was being tested
OpenAI finally dropped GPT-6.2 Astra & Sol, and both are looking really, really good across benchmarks tldr: Astra is almost on par with Fable 5.2, falling just a couple of points short on the overall intelligence index, but significantly cheaper to use (like 6x cheaper) for
something is definitely happening inside the NSA campus
the time between a security patch being released and attackers having a working exploit could soon shrink to minutes, thanks to llms
interesting theory / thread on why we haven’t seen gpt 6.2 released yet tldr, pretty much the same as the rumors going around, major security concerns and wanting to get it right without the government immediately shutting the model down
one of the craziest hardware hacks i’ve seen in a while, a security researcher just found a way to unlock the A100 capabilities Nvidia had disabled in a crypto mining GPU Kate discovered that NVIDIA’s CMP 170HX uses the same GA100 chip as the A100, but with most of its
NVDA-2.97%
another benchmark crushed by Claude Fable 5.2, this time on MazeBench, one of the harder benchmarks for long-running AI agents basically, MazeBench drops an agent into a MASSIVE 3D world with 256 interconnected rooms and 100 hidden gems, where it has to explore, manipulate
from $20.2B to just $1.58B in reported holdings in a single quarter pretty interesting turn of events for Leopold Aschenbrenner, the former OpenAI researcher who founded Situational Awareness (an ai-focused hedge fund) according to its latest 13f, the fund exited its common
could OpenAI's new partnership with METR be one of the reasons why GPT-6.2 is taking longer than expected? METR announced that it will independently evaluate OpenAI's largest-ever (!) RL run (alongside three other outside research groups), with pretty extensive access to many
and $NVDA shares haven’t even dropped by 4% yet, finally bullish?
NVDA-2.97%
ai doesn’t have to be a race where one company wins and everyone else loses Hugging Face and Openrouter are connecting their tools so people can run open models more easily and share what their agents did, helping other builders learn from it too, great collab!
honestly, it never stops surprising me how much cool visual stuff Fable 5.2 can do with just code, like this video
Load More