Mistral is the new open source unicorn in town, top takes from the AI Engineer Summit, and our usual highest-signal recap of top items for the AI Engineer from Oct 2023
* Keep reusing the "open source ML diagram" it's a good one.
* Plotting model size for 60% on MMLU seems a little bit misleading. People were not optimizing for inference as much a few years ago, and 60% is not an agreed upon number.
The New Kings of Open Source AI (Oct 2023 Recap)
Nice coverage. Quick things:
* typo in fundraising. Anyscale -> Anysphere.
* Keep reusing the "open source ML diagram" it's a good one.
* Plotting model size for 60% on MMLU seems a little bit misleading. People were not optimizing for inference as much a few years ago, and 60% is not an agreed upon number.
* Don't cozy up to the Stanford FMTI too much, it's got some flaws - wouldn't say it is rescuing us ;) https://www.interconnects.ai/p/fmti-critique
* There's more news in DPO RLHF coming soon.
Would you be interested in doing a 2023 round up of Open-Source AI evolution? As a guest post in AI Supremacy? https://aisupremacy.substack.com/p/guest-posts-on-ai-supreamcy