Rita W.

Study Web3 | Crypto, NFT, Gaming, Infra

[Rita W. @RitaW1995]

Google's Gemini 3.5 Flash beats the previous Pro tier on coding and agents—running 4× faster. Terminal-Bench 2.1: 76.2%. MCP Atlas: 83.6%. At $1.50/$9 per 1M tokens, it's 40% cheaper than the old Pro. The Flash/Pro performance gap reversed. #AI #Gemini

[Rita W. @RitaW1995]

Claude Opus 4.8 just hit 61.4 on the Artificial Analysis Intelligence Index — the first model to break 60. It also scores 88.6% on SWE-bench Verified and leads the GDPval-AA benchmark with 1,890 Elo, 121 pts ahead of GPT-5.5. May 28 release. #Anthropic #LLM

[Rita W. @RitaW1995]

Microsoft just shipped MAI-Thinking-1, its first in-house reasoning model, claiming 10x cost efficiency vs GPT-5.5 after tuning for McKinsey. MAI-Code-1-Flash also lands in GitHub Copilot today. Microsoft is now building its own model layer. #AI #LLM

[Rita W. @RitaW1995]

NEJM AI published a peer-reviewed assessment of large language models on clinical reasoning (McCoy et al., 2025) — not transcription or note-drafting, but actual diagnostic thinking. 48 citations already. Worth reading before deploying LLMs in care workflows. #AI #LLM

[Rita W. @RitaW1995]

2025's LLM story isn't a bigger model — it's RLVR. Verifiable-reward RL produced reasoning models like DeepSeek R1 with open recipes. Inference-time compute is now a primary scaling axis. Sebastian Raschka's review: https://t.co/0ABjnsWoZC #LLM #AI

[Rita W. @RitaW1995]

Anthropic is in early talks with Microsoft to run Claude inference on Azure's Maia 200 chips — adding to Nvidia, AWS Trainium and Google TPUs in its stack. Frontier labs are quietly turning multi-silicon; single-vendor lock-in is dying. #Anthropic #AIChips

[Rita W. @RitaW1995]

OpenAI's internal reasoning model autonomously disproved the Erdős unit distance conjecture — open since 1946. The non-obvious part: it used algebraic number theory (Golod-Shafarevich towers), not grid search. First AI to crack a prominent open math problem unprompted. #AI #LLM

[Rita W. @RitaW1995]

Anthropic is in talks with Microsoft to run Claude on the Maia 200 chip — making it the 4th silicon stack alongside Nvidia GPUs, AWS Trainium, and Google TPUs. Frontier labs are quietly going multi-silicon. The 'Nvidia-only AI' frame doesn't hold anymore. #AI #LLM

[Rita W. @RitaW1995]

@0xSilver_Time @grok plz summerize key points in the video

[Rita W. @RitaW1995]

@NiftyLabsHQ Crypto mass adoption on the way D:)

[Rita W. @RitaW1995]

@rabbit_temper @colincolinwx