Subscribe
Sign in
Home
Chat
Archive
About
New
Top
Discussion
Evaluating and uncovering open LLMs
When choosing a model, we're stuck in the middle between classic NLP benchmarks (e.g. MMLU) and qualitative chatbot ranking. Neither are exactly what we…
May 31
•
Nathan Lambert
6
2
Share this post
Evaluating and uncovering open LLMs
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
Code: green pastures for LLMs
Both code-generation and code-training seem central to LLM progress, present and future.
May 25
•
Nathan Lambert
9
Share this post
Code: green pastures for LLMs
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
Unfortunately, OpenAI and Google have moats
While everyone went crazy over a leaked memo, no one took the time to think through how companies have worked in the internet era.
May 17
•
Nathan Lambert
23
3
Share this post
Unfortunately, OpenAI and Google have moats
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
Specifying hallucinations
The future of LLM errors and the need for a discourse around the specification of unexpected outputs.
May 3
•
Nathan Lambert
8
6
Share this post
Specifying hallucinations
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
April 2023
Beyond human data: RLAIF needs a rebrand
Reinforcement learning from computational feedback: how a promising method is being ignored because of a confusing launch.
Apr 26
•
Nathan Lambert
13
3
Share this post
Beyond human data: RLAIF needs a rebrand
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
The next chapter: Interconnects
Democratizing Automation is now Interconnects. Doubling down on making the most important links in AI.
Apr 19
•
Nathan Lambert
9
Share this post
The next chapter: Interconnects
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
Growing needs for accessing state-of-the-art reward models
The rewards assigned by these models will control the subjective experience of users; the subjective experience controls their decision-making.
Apr 12
•
Nathan Lambert
4
8
Share this post
Growing needs for accessing state-of-the-art reward models
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
Behind the curtain: what it feels like to work in AI right now
Fear, FOMO, and the scientific exodus driven by ChatGPT
Apr 5
•
Nathan Lambert
88
21
Share this post
Behind the curtain: what it feels like to work in AI right now
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
March 2023
The implicit dynamics of optimizing costs vs. rewards vs. preferences
With the emergence of reinforcement learning from human feedback, we've been applying old techniques with a new guiding function (🤫 RLHF).
Mar 27
•
Nathan Lambert
11
Share this post
The implicit dynamics of optimizing costs vs. rewards vs. preferences
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
GPT4: The quiet parts and the state of ML
Checking in on the state of the ML field: technical progress, societal implications, and more.
Mar 20
•
Nathan Lambert
13
1
Share this post
GPT4: The quiet parts and the state of ML
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
AGI Roundup: Re-visiting Go; transitioning from narrow to general; multimodality & GPT4
AI is much more fun to think about when AGI is an adjective, not a finish line.
Mar 9
•
Nathan Lambert
2
Share this post
AGI Roundup: Re-visiting Go; transitioning from narrow to general; multimodality & GPT4
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
February 2023
The RLHF battle lines are drawn
Tales of the open and closed sides, how these two dynamics will dictate progress and public perception.
Feb 27
•
Nathan Lambert
19
Share this post
The RLHF battle lines are drawn
www.interconnects.ai
Copy link
Linkedin
Facebook
Email
Notes
This site requires JavaScript to run correctly. Please
turn on JavaScript
or unblock scripts