5 Steps To Dramatically Reduce Risks In AI Development
#AI #AIio #BigData #ML #NLU #Futureofwork @lexfridman @sama @kaifulee @ID_AA_Carmack @karpathy @2morrowknight @ylecun http://
ow.ly/5lXo30sK3lq
SAFETY
-
5 Steps To Dramatically Reduce Risks In AI Development
By
–
-
Recognizing AGI: Knowing When Tasks Are Complete
By
–
Once you truly feel the AGI you’ll just ~know~ when each task is done
-
ASI Discontinuity: Entity Distinction and Inter-ASI Negotiation
By
–
In the LDT case, the discontinuity is that ASI is (a) a whole different entity and (b) also has a whole different means of negotiation with other ASIs.
-
ASIs Don’t Need the Meta-Golden Rule
By
–
ASIs don't need the "Meta-Golden Rule" sometimes postulated by people who don't understand how the asymmetrical handshake works. The Meta-Golden Rule, as it was invented by people who didn't understand decision theory, has obvious problems from the perspective of people who do.
-
Logical Counterfactuals and Q* Correlation Requirements
By
–
If you cannot formally write down what a logical counterfactual is, and I can't, there is no way that your Q*'s expectations counterfactually correlate with Q. Being able to write down the gross formula correctly is a necessary though not sufficient condition.
-
Microsoft Quantum Results Face Significant Reproducibility Critique
By
–
Significant critique of Microsoft Quantum's results just released on arXiv: – Protocol definition differs between published paper and code – Results depend on unexplained measurement choices, not device properties Trending on alphaXiv
-
Maintaining Balance While Building AI Products for Longevity
By
–
Depends on the definition of breakthrough and balance of course but as one data point, feels like we always managed to maintain some good balance while building HF (which IMO might explain our longevity)
-
OpenAI Introduces GPT-4.5 with Enhanced Safety and Alignment
By
–
2). GPT-4.5 OpenAI introduces GPT-4.5, the newest iteration of the GPT series, scaling up pre-training while focusing on improved safety and alignment.
-
Top AI Papers of the Week: GPT-4.5, Claude 3.7, and More
By
–
Here are the top AI Papers of the Week (Feb 24 – Mar 2): – GPT-4.5
– PlanGEN
– Protein LLMs
– Chain-of-Draft
– Claude 3.7 Sonnet
– Emergent Misalignment Read on for more: -
Anthropic releases Claude 3.7 Sonnet with extended thinking mode
By
–
1). Claude 3.7 Sonnet Anthropic releases a system card for its latest hybrid reasoning model, Claude 3.7 Sonnet, detailing safety measures, evaluations, and a new "extended thinking" mode.