- Bay Area Times
- Posts
- Google’s open weights Gemma 2 nears Microsoft’s Phi-3 on benchmarks
Google’s open weights Gemma 2 nears Microsoft’s Phi-3 on benchmarks
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/e17c23f3-58a2-4d15-b965-192be2687e69/Bay_Area_Times_email.png)
Top stories today:
- Google’s open weights Gemma 2 nears Microsoft’s Phi-3 on benchmarks
- Q1 GDP revised up from 1.3% to 1.4%
- OpenAI announces CriticGPT, model to help human trainers in RLHF
- ChatGPT web visits +16% MoM, +96% YoY to 2.9B in June
0. Data and calendar
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/2dd652a9-3699-4f08-941d-ea6322eb11c1/Screenshot_2024-06-28_at_11.10.49_AM.png?t=1719569488)
All values as of 6 AM ET / 3 AM PT, other than S&P500 and NASDAQ close (4 PM ET / 1 PM PT).
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/7cc6b321-8787-4561-a1d7-e4854c6d2ed1/Screenshot_2024-06-28_at_11.33.00_AM.png?t=1719570847)
All times are ET.
Correction: We reported yesterday that Amazon's +4% to $2T market cap was based on no news. In fact, the rise was connected to the report that Amazon plans to launch a Temu-like store. h/t to our reader FD.
1. Google’s open weights Gemma 2 nears Microsoft’s Phi-3 on benchmarks
Google curiously didn’t explicitly compare Gemma 2 with Phi-3 in their launch post:
Gemma 2 runs on a single NVIDIA H100.
Phi-3 continues to be the leading open weights small language model according to MMLU, having a 78.2 score
TBD: Human comparisons between the 2 models.
2. Q1 GDP revised up from 1.3% to 1.4%
We have been using Athyna to hire our global team.
~67% savings compared to hiring in U.S./Europe.
Pay nothing until you find the perfect match.
Get $1,000 OFF on your next offshore employee by mentioning us.
*Disclaimer: Athyna is a partner of the Bay Area Times.
4. OpenAI announces CriticGPT, model to help human trainers spot mistakes during RLHF
Humans + CriticGPT provided more comprehensive critiques and hallucinated less often than just humans: