Interesting paper argues that scale is all you need to explain humanity’s unique levels of intelligence. “We propose that global, genetic differences in learning and memory are sufficient to account for uniquely human reasoning across domains,” https://
nature.com/articles/s4415
9-024-00283-3
…
@emollick
-
Scale Explains Human Intelligence Across Domains Study
By
–
-
GPT-4 Prompts Reshape Academic Recommendation Letters
By
–
The future happened when a student asking for a letter of recommendation sent me the suggested GPT-4 prompt he wanted me to use to generate the letter itself. Likely where recommendation letters are eventually going to go: 1-10 scale & a single sentence assessment. No ceremony.
-
Automating Management Writing: Executive Blind Spot on AI Impact
By
–
I have talked to a lot of executives buying Microsoft Copilot for their firms. Not one seems to have considered what it means to suddenly automate the vast majority of management writing without training or reconsidering the meaning of the work. From my book, Co-Intelligence:
-
LLM Bias in HR: Papers Reveal Hidden Discrimination Risks
By
–
This is a useful thread of recent papers showing that LLMs are biased when judging people in HR contexts. The papers offer some potential solutions, but it is important to recognize that current systems carry subtle biases that can be hard to recognize without extensive testing.
-
East Coast AI pragmatism overlooks risks of powerful systems
By
–
I generally think the East Coast business attitude towards AI is healthy – I would characterize it as a focus on "what it can do" over "what does it all mean" – but it means that most large companies don't take the risk/possibilities of significantly more powerful AI seriously.
-
Agents: Beyond Chatbots and Copilots for AI Interaction
By
–
Agents represent the first break away from the chatbot and copilot models for interacting with AI. There is something different about assigning a task to an agent like Devin, and then not worrying about it because you know it is on it & will message you if it has any questions.
-
Labs Take AGI Goals Seriously Despite Market Skepticism
By
–
I agree with this very much. But folks don't take the labs seriously when they say they are aiming for AGI and soon. Altman can tell people the goal of OpenAI is near-term superhuman computers & people assume it is marketing. It may not be possible, but the labs believe it is.
-
LLMs Struggle with Word Games Due to Token-Based Processing
By
–
They work in tokens, not letters, so are terrible at word games.
-
Academics Should Re-test LLM Performance Across Domains
By
–
With Claude 3 APIs available, I would love to see more academics re-running their tests (like the papers on AIs performance in medicine & law). Advantages:
1) Helps us understand the variance among GPT-4 class models
2) Helps us understand if there are universal LLM shortcomings -
WSJ Book Review Generated in Multiple Musical Styles via AI
By
–
Part of the Wall Street Journal’s review of my book as:
Late 90s ska: https://
app.suno.ai/song/cdf84842-
ef12-4a46-b694-a0dbfd70916c
…
1920s Victrola recording: https://
app.suno.ai/song/2db23835-
ebc2-4ae9-a537-f6d286456a26
…
Neutral Milk Hotel-ish psychedelic folk: https://
app.suno.ai/song/7f4f89e6-
b43d-4dfe-a60e-33065ee7321a
…
Europop: https://
app.suno.ai/song/4087ea84-
6686-4857-80e1-99f254c6a68c
…