Brought to you by:
Top stories today:
- Meta unveils Orion as its first AR glasses
- OpenAI CTO Mira Murati leaving, along with other 2 top researchers
- Llama 3.2 open weights LLMs launched by Meta
- X releases first transparency report since Elon’s takeover in 2022
0. Data and calendar
All values as of 6 AM ET / 3 AM PT, other than S&P500 and NASDAQ close (4 PM ET / 1 PM PT).
All times are ET.
1. Meta unveils Orion as its first AR glasses, but they won’t be for sale any time soon
“We’re giving Meta employees and select external audiences access to Orion.”
A 70-degree field of view is offered by the Meta AR glasses.
The Orion glasses are paired with a “neural wristband” and a wireless compute puck
Pipedrive’s easy and effective CRM lets you track your sales pipeline, manage deals, and automate the whole process so you can focus on selling.
Draft quick, compelling, and personalized emails from simple prompts in just a click.
Summarize threads with Al to save valuable time.
*Sponsored.
Meta also updated the Ray-Ban Meta smart glasses with real-time AI video processing and live translation
Special edition Ray-Ban Meta glasses with transparent look and transitional lenses were also unveiled:
Finally, Meta launched Quest 3S at $299.99 with Quest 3 hardware and a higher battery life of 2.5 hours
2. OpenAI CTO Mira Murati leaving, along with other 2 top researchers
Stepping away to “create the time and space for own exploration:” Murati.
OpenAI research chief Bob McGrew and research VP Barret Zoph are departing.
Mark Chen will now lead the research team at OpenAI.
Murati, McGrew, and Zoph decided to leave OpenAI independently of each other, CEO Sam Altman said
Altman reportedly could get 7% of the new for-profit OpenAI, a $10B equity grant
OpenAI is also planning to unveil a new version of Sora, its text-to-video model, but it is unclear whether it will be released.
3. Llama 3.2 launched by Meta: 11B, 90B multimodal; 1B, 3B text-only open-weight models
Llama 3.2 90B beats GPT4o-mini and Claude 3 Haiku on major benchmarks, according to Meta:
The 90B model is easier to deploy than Llama 3.1 405B, as it can run in a single system with 8 x H100s.
Mistral’s Pixtral 12B, however, overperforms Llama 3.2 11B on most benchmarks.
The smaller 1B and 3B models are on par with Google’s Gemma 2 2B and Microsoft’s Phi-3.5-mini, also open weights
128K context length.
On-device focus.
Optimized for Arm hardware with Qualcomm and MediaTek chips.
The new smaller Llama 3.2 models are among the cheapest, on par with Gemini 1.5 Flash and Pixtral
4. X releases first transparency report since Elon’s takeover in 2022
There were 224M user reports in H1, with 36%+ of them for abuse and harassment, followed by ~30% for hateful conduct and ~18% for violent content:
