When given screenshots of an unreleased game that was just revealed this week (that I am very much looking forward to), GPT-4 is able to make some pretty impressive educated guesses, including the game's title, without access to the internet.
@emollick
-
Starting with AI: The case for persona-based interaction
By
–
Thats not really my argument, for what its worth. I argue that for most people, talking to the AI is the place to start, especially after giving it a persona.
-
LLM-Powered Eldritch Site Generates Six Unique Experiences
By
–
The first draft of the Eldritch site is done, it created six different experiences (all based on the LLMs judgement, I gave it no suggestions). I gave it some feedback to implement, but it unexpectedly created a custom 404 error page, which is a very unexpected but nifty touch.
-
LLMs Demonstrate Unexpected In-Context Regression Capabilities
By
–
What??? “Our results demonstrate that large language models are capable of doing regression when given in-context examples of (input, output) pairs, despite not being explicitly trained to do so.”
-
Developer launches 1980s BBS simulator project
By
–
Great, now I have to do work for it. (I started another project, a 1980s BBS simulator).
-
Devin Implements Safety Controls for Eldritch Website System
By
–
The fact that Devin has implemented both a "Madness Meter" and "ForbiddenKnowledge.js" for the Eldritch website is promising
-
AI System Self-Debugging and Quality Assurance Capabilities
By
–
Debugging its own software by running what it creates and doing some QA. Nifty.
-
Canvas AI Documentation Lookup Feature Implemented
By
–
Neat, it is looking up documentation from Canvas!
-
Devin AI Agent Performance on Eldritch Site Grader Tool
By
–
I'll let you know how the Eldritch site and grader tool work out. I have no idea how well Devin will do on them. Exploring the jagged frontier!
-
AI Agents Transform Task Automation With Minimal User Effort
By
–
I know I keep tweeting about AI agents, but using one feels so much different than the effortful work in using a chatbot. You can just assign it tasks & check back later to see the results. Sometimes it amazes, sometimes it needs feedback or is meh. But it doesn't require effort