Mastodon discussion Discussions Sep 30 1 views

what i'd actually do this week:1. put gpt-6.1 sol and sonnet 5.5 on the same eval set2. compare cost per task, not per t...

by Mohith kumar

what i'd actually do this week:1. put gpt-6.1 sol and sonnet 5.5 on the same eval set2. compare cost per task, not per token3. keep astra and opus only for what the cheap ones fail4. skip ultrafast unless a human is waiting on itwhich one wins your evals?source: https://openai.com/index/introducing-gpt-6-1-sol/#AI #OpenAI #Anthropic #LLM

Read Original

Anthropic Benchmark OpenAI

Metadata

Favourites Count: 1
Account: mohith808

Mastodon discussion 16m ago

What makes this paper stand out is its focus on tactile feedback for dexterous manipulation — a step toward robots that ...

What makes this paper stand out is its focus on tactile feedback for dexterous manipulation — a step toward robots that can handle objects as subtly as humans do.https://huggingfac...

Mastodon discussion 17m ago

Google unveiled two new TPU variants: TPU 8tx, built with AMD for confidential AI training and inference, and TPU 8i Max...

Google unveiled two new TPU variants: TPU 8tx, built with AMD for confidential AI training and inference, and TPU 8i Max, which doubles memory, compute and bandwidth. It's part of ...

Mastodon discussion 17m ago

what i'd actually do this week:1. put gpt-6.1 sol and sonnet 5.5 on the same eval set2. compare cost per task, not per t...

Metadata

Related

What makes this paper stand out is its focus on tactile feedback for dexterous manipulation — a step toward robots that ...

Google unveiled two new TPU variants: TPU 8tx, built with AMD for confidential AI training and inference, and TPU 8i Max...

fixedIt https://www.reddit.com/r/ProgrammerHumor/comments/1xsngn9/fixedit/#ai