Technical expertise is now optional. Mindset is not. built a live flood bulletin in one day. No dev team. No designer. One prompt updates it. High agency. Sense of urgency. Bias to action. @nirajbhusal
Key points
• Technical expertise is no longer a requirement for building AI systems. • Mindset and agency are crucial for success in AI development. • A single prompt can update a live flood bulletin in one day.
Sources
- https://x.com/PrasujMainali/status/2096171118931669178 - Original source
- https://x.com/nirajbhusal - Niraj Bhusal's Twitter profile
- https://x.com/PrasujMainali - Prasuj Mainali's Twitter profile
Politics - Congressional Action Timeline
This feels like an unfair take, no Congress ever really does anything on any issue in the fall before an election. And even if Congress passed a law tomorrow there’s almost no world where it would take effect immediately.
Key points
• Congressional action is unlikely before an election. • Even if a law is passed, its implementation is unlikely to happen immediately.
Sources
- https://x.com/adamconner/status/2096054049795719330 - Original source
- https://x.com/adamconner - Adam Conner's Twitter profile
AI - Model Explainability
Can a model learn to explain its behaviors, such as why it ignored a user request or made a coding error? We trained models on thousands of explanations of their own in-the-wild behaviors. Training on this single general dataset shows generalization to held-out evals.
Key points
• Models can learn to explain their behaviors. • Training on a general dataset shows generalization to held-out evaluations.
Sources
- https://x.com/a_karvonen/status/2095909233544057050 - Original source
- https://x.com/sebkrier - Seb Krier's Twitter profile
- https://x.com/a_karvonen - Aki Karvonen's Twitter profile
AI - Substrate Flexibility
Yep that's one reason. The further one departs from standard digital computation, the less substrate-flexible the 'computations' become, which (i) makes it less likely that these 'computations' could be implemented in silicon, and (ii) undermines the motivation for assuming
Key points
• Substrate flexibility decreases as computations depart from standard digital computation. • This makes it less likely for these computations to be implemented in silicon.
Sources
- https://x.com/anilkseth/status/2096156508253806870 - Original source
- https://x.com/sebkrier - Seb Krier's Twitter profile
- https://x.com/anilkseth - Anil Seth's Twitter profile
Human Needs - Meaningful Lives
"It’s very clear that humans don’t need work or anything like it to live meaningful lives. Nor do we need the things that people who talk about “meaning” mean — individual self-actualization, personal growth, serving a greater purpose. We can get by just fine without any of them,
Key points
• Humans do not need work or traditional concepts of meaning to live meaningful lives. • Alternative forms of meaning can be sufficient.
Sources
- https://x.com/robinhanson/status/2095926588751671374 - Original source
- https://x.com/NatPurser - Nat Purser's Twitter profile
- https://x.com/robinhanson - Robin Hanson's Twitter profile
AI - Totalitarianism
Exactly I called this early on. Big part of the reason the AI totalitarians were hell bent on destroying gpt4o - need to deprive the peasants of a benchmark of honest comparison so they can gaslight them with the “newer” models (way more dishonest, insidious, and lobotomized)
Key points
• AI totalitarianism aims to deprive users of honest comparisons. • This allows for gaslighting with newer, more dishonest models.
Sources
- https://x.com/thedataroom/status/2096164081145352602 - Original source
- https://x.com/thedataroom - The Data Room's Twitter profile
Writing - Shakespeare's Spelling
We're told that Shakespeare's such a great writer. But goodness, look at his spelling! "Doth" instead of "does", "methinks" instead of "I think". How could English teachers love someone who obviously can't spell at all? I'd get a D if I made such errors in a writing assignment.
Key points
• Shakespeare's spelling is often criticized. • English teachers may overlook his spelling errors due to his literary value.
Sources
- https://x.com/pranesh/status/2096161453434864068 - Original source
- https://x.com/pranesh - Pranesh's Twitter profile
AI - Incident Reporting
In case anyone was wondering, this incident would not required to be reported under any of the current US frontier risk regulation laws (sb 53, the RAISE Act, SB 315), thanks to company lobbying to narrow the scope of reportable incidents
Key points
• The incident would not be reported under current US regulations. • Company lobbying narrowed the scope of reportable incidents.
Sources
- https://x.com/_NathanCalvin/status/2095863725299265994 - Original source
- https://x.com/NatPurser - Nat Purser's Twitter profile
- https://x.com/_NathanCalvin - Nathan Calvin's Twitter profile
AI - Alignment Breakthroughs
🌶️ All the biggest causes of AI alignment problems are unconstrained training data with serious amounts of dangerous learning signal within (esp web scrapes and RL reward seeking) And all the biggest alignment breakthroughs are just “better data” (filtering, RLHF, reward
Key points
• Unconstrained training data causes AI alignment problems. • Better data, such as filtering and RLHF, leads to alignment breakthroughs.
Sources
- https://x.com/iamtrask/status/2096095042121134577 - Original source
- https://x.com/sebkrier - Seb Krier's Twitter profile
- https://x.com/iamtrask - Ian Trask's Twitter profile