In this essay, @arjun_ramani3 and @zhengdongwang assemble the best arguments that they have encountered for why transformative AI is hard to achieve… 1) The transformational potential of AI is constrained by its hardest problems 2) Despite rapid progress in some AI
@nathanbenaich
-
AI Analyzes Company Success Factors Like Experienced VCs
By
–
it was actually fine – I just asked "what are they key reasons why company X (from the pdf I uploaded) would succeed or fail" and it generated totally reasonable Qs that a VC takes 10 years to learn 🙂
-
LLM Hallucination: Document Analysis vs Factual Accuracy
By
–
it is quite amazing how LLMs like claude v2 can in one turn give you solid responses about a document i uploaded and in the next turn when i ask about a specific person, the response is entirely made up. –> further to @bxchen recent piece, LLMs are useful when constrained to a
-
LLM Overkill: Do We Really Need AI for Robot Vision Tasks?
By
–
while cool, I’m not sure you need an LLM to identify an extinct animal in an image, translate it’s coordinates to a robot arm that can then pick it up.
-
Cyber and Dark Web Monitoring Software for Organizations
By
–
tbh this finding makes the case more for cyber / dark web monitoring software so orgs can realise what expert information actually lives out there on the web.
-
Are Large Model Scaling Laws Becoming Less Predictable?
By
–
no more like large model scaling laws – if they keep changing these less law-like and more hypothesis like?
-
Scaling Laws Should Be Called Scaling Hypotheses Instead
By
–
feels like scaling “laws” should perhaps be scaling “hypotheses”
-
Corporate AI Investment: Amazon, Stripe, Google Compared
By
–
To play devil’s advocate, by that measure, Amazon would have been the best corporate investor, so would Stripe, so would Google etc etc.
-
AI Applications for Longevity and Health Improvement
By
–
I guess you’re not interested in living a longer and healthier life. That’s cool!
-
Synthetic Coevolution Platform Advances Protein Interaction Analysis
By
–
“The integration of a synthetic coevolution platform with machine learning has allowed us to interrogate a protein-protein interaction with exceptional granularity, but also to use this information prospectively. Using this approach, it is possible to revisit basic principles