not worse than aviation, automobiles etc by any stretch. there should be some regulatory burden since there are clear risks that the tech companies themselves have acknowledged.
SAFETY
-
Can AI Models Create New Knowledge? Attribution Matters
By
–
Yeah like a fancy but lossy database. It’s a good question if these models can even create new knowledge? Either way I feel like attribution is so key — and at least Google search or perplexity has that.
-
Claude’s Excessive Refusal Triggers Spark Debate
By
–
Claude has the most insane refusals – definitely over triggers often for the most innocuous stuff
-
When will we achieve the first artificial general intelligence?
By
–
When will we get the first AGI?
-
Grim Implications Unrelated to Separate ASI Extinction Risk
By
–
Grim if true (for reasons basically unrelated to the totally separate track where later ASI later kills everyone later)
-
AGI before 2040: acknowledging LLM limits and next steps
By
–
AGI before 2040 could definitely happen. There are a lot of problems to solve but a lot of money and people working on it. The sooner people acknowledge the limits on LLMs, the sooner we might move on to what happens next. And it is hard to predict that. See also my fun dialog
-
LLMs as Lookup Tables: Limitations in Reasoning
By
–
Easily one of my favorites and hilarious how elaborate the thinking can be while getting it badly wrong despite clear instructions. Always reminds me these LLMs are still more like fancy and lossy lookup tables than reasoning engines
-
LLMs Lack Emergent Reasoning, Success Is Lucky
By
–
Many people are trying to make excuses for why this trivial problem trips up LLMs. Occam's razor should remind you that there is no emergent reasoning in LLMs. Instead it gets lucky (and it is astonishing that it works so well–that says something deep about our language)
-
Companies Must Comply With Existing AI Law Not Circumvent It
By
–
It‘s not more legislation that we need, but companies not trying to work around the existing law. More enforcement is needed.
-
Perplexity AI Accuracy: Can It Verify Self-Reported Claims?
By
–
In the @lexfridman podcast, he asked me whether him saying (on the podcast) that he's smart and handsome (and making the transcript available on the web) would make perplexity say that he is. The answer here is quite impressive. It caveats it with the background context that this