AI Top-of-Mind for 5.31.24 — xAI
Today: xAI funding, Phi-3 Vision capabilities, parallels to the enclosure movement, and detection engineering
Top-of-mind is confirmation of xAI’s $6 billion round at an $18B valuation. An advantage of xAI is that its developments span Musk’s other companies including Tesla, SpaceX, robotics, and implants.
From the ‘VentureBeat’ article:
· In March this year, the company again started shifting gears. It first made Grok-1 open-source and then launched its better-performing successors — Grok-1.5and Grok-1.5V.
· Grok-1.5 introduced enhanced reasoning and problem-solving capabilities and closed in on the performance of known open and closed LLMs, including GPT-4 and Anthropic’s Claude 3.
Turning to models, a look at Phi-3 Vision MLLM. Disclaimer — I’m not too familiar with this one. Youness Mansar in ‘Towards Data Science’ looks at some cool features while running on a local machine and describes the installation process. Some of his examples:
- Optical Character Recognition (OCR)
- Image Captioning
- Table Parsing
- Figure Understanding
- Reading Comprehension on Scanned Documents
- Set-of-Mark Prompting
Input:
Output:
This image shows a plush toy resembling a character from a popular animated television series. The toy has a round body with a hood, and it is wearing a pair of glasses. The character’s face is not visible in the image.
Also on the model front, an in-depth and very technical posing by Devansh on ‘Angular Embeddings’ and their impact on LLM development.
Turning to employment, some good thinking on previous societal transitions across history and what that meant for the workers. Hint — nothing good! Amanda Claypool looks at the enclosure movement in England and draws parallels to the industrial revolution and now AI. Some takeaways:
· The lessons of land enclosure in England are important to understanding the distribution of productive assets in today’s economy. Those assets — algorithms, digital platforms, and data — are used by all but owned by just a few.
· Like the peasants of Europe, today’s workers — ranging from esteemed lawyers in Boston to garbage pickers in the slums of New Delhi — will be left equally dispossessed by the new economic system that will inevitably replace the consumer-capitalist system we know today.
Lastly, on security, Dylan demonstrates how to leverage Gen AI to automate detection writing. As background on detection engineering, a link to Datadog.