Loading activity…
@nazneenrajani
ceo @collinearAI
AI
The reported MATH-500 score rose from 7.9% to 48.1% with the same sampling settings.
AI
A post says Carbon scored 88.2% on SEC-bench Pro and is rolling out first to cybersecurity specialists.
AI
Its developer says Flint exceeds 125 tokens per second on an M4 Pro MacBook Pro at short contexts.
XAI
In the announcement, @bneyshabur places Mirendil-Base in the 100B–500B-parameter range and promises independent evaluations.
AI
One post argues raw-data brokerage will commoditize, while refined post-training tasks will grow more differentiated.
AI
A post frames Blackstone's reported move as part of a broader enterprise push to build software factories.
XAI
A post argues that human tasks are now too easy to use as AI benchmarks in many domains.
AI
A commentator says new entrants can reach $80M–$100M in one-time revenue from one account without prioritizing margins.
AI
The New York Times reports the startup trains AI on materials and designs to build tiny physical devices.
AI
A CollinearAI-affiliated poster also invited people interested in RSI, post-training or cyber to DM about an invite-only lunch.
AI
A user quote-posted another account’s post with only a salute emoji as commentary.
AI
An October 1 post pairs CWE-bench v1 with Monday and FrontierCUA with Wednesday, praising the team’s pace.
AI
Google announces a 1 million-token output limit, up from 64,000, for its new model. Initial access goes to trusted cyber defenders through the Fairwind Program.
AI
A post invites people to the author's HQ office for discussions with researchers from Frontier Labs training cyber models. It also mentions a live dosa station, lassi and pani puri.
AI
Artificial Analysis says Grok 4.7 (xhigh) and MiMo-V2.6-Pro tied for the top index score of 56, while safety refusals held back several other models.
AI
One post argues that founders tweeting all day can signal fundraising and “aura-farming” aimed at justifying valuations unsupported by revenue growth.
AI
A Haize Labs co-founder announced the acquisition and said he is joining Beacon as VP of AI Research, aiming to bring trustworthy AI to essential Main Street businesses.
AI
SemiAnalysis says the two models are comparable to GPT-6 and Fable 5.1 on Terminal Bench 2.1, but their Terminal Bench 4.0 performance is markedly worse.
AI
Collinear AI CEO builds on prior research for AGI cybersecurity evaluations.
AI
Nazneen Rajani announces benchmark of tasks testing coding agents on real code defense.
