Dhaval (@SiftByte) header
Dhaval (@SiftByte) profile picture

AI / Tech Ready to learn and debate things respectfully. Views personal.

Joined September 2026
Team: Anthropic is about to go public, but we have been making a lot of progress on models Sundar: Release them Team: Which ones? Sundar: Everything
10
3
17
704
This guy predicted Fable 5.2 and GPT-6.2 Astra and Sol release before anyone, surprisingly he has only 2 post. did any get to know who th is @icedritual ?
6
2
16
469
🚨 GPT-6.2 Astra helped design the Chips at AMD. Apparently, OpenAI and AMD were working together on it, and it seems that the models are doing really well at certain chip design tasks. This is a huge leap, if you know how hard it is to design chips and build them from scratch. If we take @thsottiaux word for it, then GPT-6.2 Astra is really, really different, a different beast altogether for the hardest tasks. If it can help in some way to design a chip, then GPT-6.2 Astra is a really cracked model at everything. teapawt.org/thsottiaux/status/2123…
GPT-6.2 Astra is pretty good at hard things. Here the team shared some results on AMD and Jalapeño chip optimizations. I do get the feedback on its communication style and we're working on that, but for really the hardest tasks in the world it does seem to be in a league of its own at the moment.
2
2
9
541
🚨 Google just released the Gemini 4 Neon, and holy shit, it's beating Opus 5.5 and GPT-Sol 6.2 in every parameter out there by a large margin. Anthropic is here to catch up, and yes this is not fake, its real and verified bechmark, now waiting for the Opus 5.6 to drop. This is getting much more interesting; Google is giving a tough fight to Anthropic, which no one would have thought just 2 months back. Never bet against Google.
🚨 Gemini 4 Neon's benchmark is floating around. It's beating Google's best model, Carbon, by a fair margin and leads on 7 out of 8 parameters it was tested on. These are unverified and rumoured benchmarks, so take them with a pinch of salt.
2
1
1
14
448
🚨 OpenAI called GPT-6.2 Astra its first externally deployed model to hit the High threshold for AI self-improvement. High is roughly equivalent to giving every OpenAI researcher an extra strong research engineer. They tested it with five research-engineering tasks in two stages. Clearing either first-stage test triggers the second stage. Astra cleared only one of the three second-stage cutoffs and missed the other two by a wide margin. It saved 42.87% of training time on NanoGPT (vs 80% cutoff) and closed 74.79% of the gap on PostTrainBench Lite (vs 90% cutoff). Sol stayed below High entirely. OpenAI is calling this High. The numbers show it still fell short on two of the three harder research-engineering tests.
5
1
11
515
🚨 Gemini 4 Neon's benchmark is floating around. It's beating Google's best model, Carbon, by a fair margin and leads on 7 out of 8 parameters it was tested on. These are unverified and rumoured benchmarks, so take them with a pinch of salt.
8
8
4
80
9,129
Finally hearing something good here.
1
2
36