Pedro Domingos
@pmddomingos
Professor of computer science at UW and author of '2040' and 'The Master Algorithm'. Into machine learning, AI, and anything that makes me curious.
My new TED talk on how AI will exponentially increase our collective intelligence: go.ted.com/zKKbE
The beauty of benchmarks is that there are so many you can always find one your model is good on.
In 100 years Europe went from ruling the seas to being at the mercy of the Houthis.
The universe is still in beta. In the final release the bug that allowed humans to emerge will be fixed.
The smarter your AI is, the less data it needs. Needing huge training sets is a sign of stupidity.
The Houthis, a no-good rebel group, sink European ships at will and fought the US Navy to a standstill. Let that sink in.
The techtonic plates are shifting. The Nvidia plate is rising, and the Apple plate is subducting.
You know a subfield of AI is running out of options when it turns to reinforcement learning.
A long long time ago (months) Google was the one notorious for paying insane amounts for tech talent.
Forget coding and math reasoning. AI has already mastered the most important skill.
🤔 Feel like your AI is bullshitting you? It’s not just you. 🚨 We quantified machine bullshit 💩 Turns out, aligning LLMs to be "helpful" via human feedback actually teaches them to bullshit—and Chain-of-Thought reasoning just makes it worse! 🔥 Time to rethink AI alignment.
Watch out, managers. More and more AIs are earning MBAs (Machine Bullshit Awards).
🤔 Feel like your AI is bullshitting you? It’s not just you. 🚨 We quantified machine bullshit 💩 Turns out, aligning LLMs to be "helpful" via human feedback actually teaches them to bullshit—and Chain-of-Thought reasoning just makes it worse! 🔥 Time to rethink AI alignment.
Of all the tech giants, Google has the most to gain from AI and also the most to lose. No pressure, Sundar.
Attention is all you need. Oh, and also MLPs, layer norm, resnets, positional encoding, tokenization, Adam, fine-tuning, GPUs, etc.
Transformers are the standard model of AI. They’re both: - The result of a long series of breakthroughs - Surprisingly powerful - A big mess - Missing key pieces
Nailed it.
Terence Tao on the supposed Gold from OpenAI at IMO