Moved on again -see 🦋https://bsky.app/profile/austegard.com
Formerly barely read at https://x.com/austegard, now barely read there.
| More | https://austegard.com |
| Moved to | https://bsky.app/profile/austegard.com |
Fascinating, if a bit hard to follow analysis by @cjhandmer
Solar and batteries for generic use cases
https://caseyhandmer.wordpress.com/2024/11/09/solar-and-batteries-for-generic-use-cases/
Amazing and promising trends!
PS! Claude/ChatGPT are both good at explaining the post...
"The amount of energy needed to refute bullshit is an order of magnitude bigger than that needed to produce it."
4 years of Brandolini's Law fallout -- what will be the effects?
Classic 80/20 rule: took 3 tries to get the initial working form, then 30 more tries to get it to properly implement HistoryAPI, a missing field, UX tweaks (me to discover a schema record was causing UX issues), etc, etc.
See final? version with example search url sync at https://austegard.com/llms.html?search=claude-3-5
Still, the 0 to 1 step is incredibly important and liberating!
No shocker that Claude 3.5 Haiku expertly solves the "99% to 98% blue marbles" question. A bit surprised it's not faster. Note that CoT is baked into the model, with no system prompt provided!
**The Great Acceleration**
Preparing Society for Exponential AI Evolution
As AI evolves at breakneck speed, we face an unprecedented challenge: adapting to cognitive partners that may soon rival our intelligence. This essay explores how we can consciously shape this transformation, balancing institutional responses with personal agency to ensure AI enhances rather than diminishes our humanity in the years to come.
https://gist.github.com/oaustegard/e01c45ac821cb5366d4c7731dd754cdc
OpenAI's CHatGPT Search is good, and Google should be concerned, BUT:
Its crawler violates publishers' robots.txt policies and the model is excerpting and summarizing news articles from publishers with which OpenAI does NOT have agreements (e.g. Gannett)
Is incapable of reasoning on follow up questions: it is more similar to a one-and-done search than an AI Chat
Amazingly makes use of Google's cache of each source's Favicon (presumably for UX rendering speed)
A colleague came across https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json and shared it with me, thought we could maybe convert to a row based format and use in a spreadsheet.
10 minutes later we instead had this, created in 3 tries by Claude 3.5 Sonnet (new aka 20241022). It was harder deciding how to host it than to "code" it!
https://austegard.com/llms.html
A searchable, sortable table of 647 LLM models/providers
How many other such small but previously tedious use-cases are now trivial?
