When I tweeted this in January 2023, a single share of Nvidia was worth roughly $150. Today, a single share is over $735.
AI
-
AI Agents Learn Automation From Screen Recording Examples
By
–
We'll soon be able to give AI agents a screen recording example, from which they'll be able to infer what's going on, and then create automations that can take over repetitive work.
-
Identifying Programming and Mapping Technical Books Collection
By
–
Ran this quickly on that image: === Low Detail === The books are partially obscured, but I'll list the titles that I can read or identify: 1. "Mapping Hacks" by Schuyler Erle, Rich Gibson, Jo Walsh
2. "Programming Hacks" (full title and authors not visible)
3. "The Return of -
Token Optimization and Custom AI Chips for Scaling
By
–
Tokens are a function of how optimized your tokenization system is, as well as your embeddings. Therefore, new breakthroughs can emerge that don't require an enormous amount of GPU power. Sam needs money to create his own chips and scale up more quickly. We've seen that Sora's
-
Token Efficiency in Vision AI Models Comparison
By
–
I'm impressed with the token efficiency! 258 tokens representing the image and being able to get all that detail out is really good. I recall that GPT4-vision in low quality mode does ~100ish tokens and doesn't get detail out that well. Might be an interesting case for a
-
Advanced GPT Version Suspected as AI Research Assistant Tool
By
–
Also, remember that Sora joke I made last week? After this experience, my hunch that a more advanced version of GPT is being used as a research assistant tool has just gotten stronger.
-
AI Understanding and Deducing Novel Information Effectively
By
–
I provided a long piece of information that I am sure wasn't present in the dataset, and not only was it able to understand the information but also capable of deducing elaborate solutions and ideas from it.
-
Extended Context Windows Enable Better AI Processing and Reasoning
By
–
Imagine providing lots of context and still having tokens left to elaborate and ask questions with near-perfect memory. The only drawback is that it takes a little longer to process the information, but honestly, it's a trade-off I am willing to accept for better results.
-
AI Understanding Novel Information and Deductive Reasoning
By
–
I want to package it in a way that makes sense, but the TLDR is that I provided a long piece of information that I am sure wasn't present in the dataset, and not only was it able to understand the information but also capable of deducing elaborate solutions and ideas from it.
-
Gemini Pro 1.5 Extended Context Delivers Authentic Understanding
By
–
I just had a true jaw-dropping moment using Gemini Pro 1.5. One thing is clear, this extended context really makes a difference.
Secondly, this is the closest I've felt to what seems like authentic understanding from an LLM.