Fun little Claude 3 prompt: "you are going to set up a little ASCII town with different characters, like an old school top down RPG, I will give you commands and meet people and do stuff" (Interestingly, GPT-4 is good at the adventure piece, but hallucinates a lot of the ASCII)
@emollick
-
Self-Play Training Could Overcome Human Data Limits for LLMs
By
–
Why this matters: human written data & human reinforcement may set the upper bounds on what LLMs can do. AlphaGo was able to beat humans because it trained by playing against itself. This suggests that LLMs may be able to do similar self-play, offering a path to rapid improvement
-
Apple Vision Pro Lacks Killer Apps for VR Launch Success
By
–
It is weird that, after all the investment, Apple hasn’t spent the money for more super high quality apps to support the launch. There are no social experiences at all (the killer VR feature), no games, no communal workspaces. Just the world’s most awkward Zoom implementation.
-
Four AI Futures: Plateau, Linear, Exponential, or AGI
By
–
My book (Co-Intelligence) ends with four scenarios for the future of AI: quick plateau, linear ability growth, exponential growth, or AGI. I made a GPT that will help you game out the scenarios: https://
chat.openai.com/g/g-TcSx5BYQr-
the-four-futures-planner
… You can you can buy the book here: -
Google’s 40M Books: From Orphaned Project to AI Advantage
By
–
If this is still true, Google's 40M+ books that they scanned over the past decade are going to go from an orphaned project to a surprising source of advantage.
-
GPTs Are Weak Form of Agents, Says OpenAI
By
–
GPTs are the weak form of agents. OpenAI has been pretty explicit about that.
-
Fortune 1000 Companies Deploy GPTs for Internal Enterprise Automation
By
–
Moderna has deployed 400 GPTs for things like contract checking & research. I spoke to a couple other Fortune 1000 firms that have also deployed GPTs internally. I am actually a bit surprised at how much internal experimentation is happening.
-
Llama-2 Improves Game Performance Through Adversarial Self-Play
By
–
This is worth noting: it appears to be the first evidence that LLMs can improve performance at a task through self-play. In this case, Llama-2 7B was able to greatly improve its ability to play the game of Adversarial Taboo by training through adversarial competition with itself
-
Local LLMs on Phones Connecting to Cloud Intelligence
By
–
Pretty clear signaling from the flurry of papers by Microsoft and Apple that local LLMs on phones connecting to bigger LLMs in the cloud are coming. Phi 3 is already far better than Siri or Alexa, and with tools and the ability to call on a smarter intelligence, it can do a lot
-
AI Technology Deciphers Ancient Herculaneum Scrolls
By
–
It is pretty amazing that the first 1,000 words recovered from the charred scrolls at Herculaneum taught us two new things: where Plato was buried and that he was sold as a slave by the Spartans. Usually, most data you get is boring, a promising start! https://
cnr.it/it/nota-stampa
/n-12655/lo-sguardo-tecnologico-legge-i-papiri-carbonizzati
…