After ChatGPT, the future belongs to multimodal LLMs. What’s even better? Open-sourcing.
— Jim Fan (@DrJimFan) March 7, 2023
Announcing Prismer, my team’s latest vision-language AI, empowered by domain-expert models in depth, surface normal, segmentation, etc.
No paywall. No forms. https://t.co/LV76hKH1PY… pic.twitter.com/U4NeizwGX8
There're 3 major bottlenecks for robotics: data, data, data.
— Jim Fan (@DrJimFan) April 11, 2023
Amazon ARMBench completely flew under the radar, but I believe is one of the most exciting data releases in robotics:
- Real warehouse deployment scenarios: not as sexy as ChatGPT, but holds gigantic economic… pic.twitter.com/10Ma8unOsM
- Blog post: https://t.co/YbCvLm0FRU
— Jim Fan (@DrJimFan) April 11, 2023
- Project website & download: https://t.co/jEpuUEQAHy
- ARMBench paper: https://t.co/cwqr3rU1it by Chaitanya Mitash, Fan Wang, Shiyang Lu, Vikedo Terhuja, Tyler Garaas, Felipe Polido, Manikantan Nambi
Many of you ask, why can LLM ever be superhuman if it’s trained on human data?
— Jim Fan (@DrJimFan) April 11, 2023
The reason is simple. It’s far easier to judge whether a work is good or not, than to create the work yourself.
You can be a food connoisseur but terrible at cooking. You can be a top art critic but… pic.twitter.com/lnaUVYgNjt
Below is Steve Jobs announcing a new product, “WiFi-connected socks” in his eloquent voice. I’m sold.
— Jim Fan (@DrJimFan) April 10, 2023
The Black Mirror episode “Be Right Back” describes a technology that revives people from their digital footprint. Where we are today:
- LLMs prompted by past chat logs can do a… pic.twitter.com/NEkUxR0P42
A very clean toy experiment. Next up:
— Jim Fan (@DrJimFan) April 9, 2023
- Does few-shot in-context emerge? At what model size, data size, and data diversity? Prompt with a set of new transitions.
- RL from human feedback: how many bits of human annotation are required to produce a desired distribution?
- Safety:… https://t.co/lHZEzqQxFq
Deepmind released a comprehensive overview of transformer architectures and algorithms!
— Lior⚡ (@AlphaSignalAI) April 9, 2023
This is a must-read to understand language models. It covers what they are, how they are trained, what they are used for, and their key architectural components.https://t.co/MCOYoMW8nN pic.twitter.com/P28SRn0Yrj
For every new iPhone, we expect longer battery life.
— Jim Fan (@DrJimFan) April 8, 2023
Context length is the new battery for LLMs. For every new GPT, there will be more, until context becomes infinite.
In-context retrieval is not new and can be deeply integrated with attention, instead of embedding search. pic.twitter.com/VET1vQ74QA
An enforced silence cannot solve India’s problems The government’s deep-rooted disdain for democracy and democratic accountability is disturbing ....... Faced with a determined Opposition, the Narendra Modi government resorted to unprecedented measures — expunging speeches, preventing discussion, attacking Members of Parliament and, finally, disqualifying a Congress Member of Parliament at lightning speed. As a result, the Budget of 45 lakh crore of rupees of people’s money was passed without any debate. In fact, the Prime Minister was busy inaugurating projects in his constituency with widespread media coverage when the Finance Bill was rammed through the Lok Sabha.
China Mandates Security Reviews for AI Services Like ChatGPT
The best VPNs for 2023
Meta’s New AI Can Pick Out and Cut Any Object in an Image—Even Ones It’s Never Seen Before
Physicists Recreated the Famous Double-Slit Experiment Using Time Instead of Space
No comments:
Post a Comment