Mastodawn

Jose M.Feb 4, 2024

Bret Victor Feb 1, 2024

Dynamicland's new research website will be up in the spring.

Jose M.Feb 4, 2024

Pragmatic Bookshelf 📚Feb 3, 2024

The iconic Pickaxe. 5th Ed, now in print. 716 pages, current thru Ruby 3.3.
(🤫 newsletter coupon still valid... RUBY5COMPLETE for 40% off ebook at https://pragprog.com/titles/ruby5/)
For more news and discounts, sign up for the newsletter: https://pragprog.com/newsletter/

https://pragprog.com/titles/ruby5/programming-ruby-3-3-5th-edition/

Show thread

Jose M.Dec 28, 2023

@ekuber @[email protected]
One metric I wish they could gather is what level of experience developers had and whether the same errors were minimized after becoming aware on how to avoid it. In any case, I'm glad they're spending time and effort on error observability and, of course, your work on the compiler team. Thanks

Jose M.Dec 27, 2023

@[email protected]
I wonder @ekuber if you have any opinion on these findings. Are any of these types of errors better handled by the compiler itself? Did you find any of these surprising?

Thanks.

Jose M.Jul 4, 2023

The comments in this post just tell me how accustomed we are to the most superficial use of computers.

Reminds me of Silo.

I can only imagine what we could do if we had the epiphany of what we can do with it.

https://lobste.rs/s/fvnuvm/xerox_smalltalk_80_gui_was_weird

#computing #smalltalk #programming

The Xerox Smalltalk-80 GUI Was Weird | Lobsters

Jose M.Jun 15, 2023

Don't ever trust Yelp reviews

https://www.sfgate.com/tech/article/yel-review-fraud-kay-dean-18150617.php

#yelpcantbetrusted #yelp #fakereviews

Bay Area woman on a crusade to prove Yelp reviews can't be trusted

San Francisco-based Yelp is up against a former government investigator who is on a yearslong crusade to prove that online reviews can't be trusted.

SFGATE

Jose M.May 31, 2023

And just as I wrote yesterday on using AI to fix AI issues, OpenAI releases a paper on an approach to do just that:

https://openai.com/research/improving-mathematical-reasoning-with-process-supervision

#ai #openai #aitofixai

Improving mathematical reasoning with process supervision

We've trained a model to achieve a new state-of-the-art in mathematical problem solving by rewarding each correct step of reasoning (“process supervision”) instead of simply rewarding the correct final answer (“outcome supervision”). In addition to boosting performance relative to outcome supervision, process supervision also has an important alignment benefit: it directly trains the model to produce a chain-of-thought that is endorsed by humans.

Jose M.May 30, 2023

I wish people would perceive any AI issues - take hallucinations as an example - as a problem that would be solved by another AI or more specifically, another AI algorithm.

Take ChatGPT which is really GPT + RL. Transformers are incredible but equally impressive are RLs. What would a second pass through RLs do for source verification? And that's assuming that the RL already in place - mostly done by human testers, btw - isn't enough to handle all current issues.

#AI #RL #GPT #ChatGPT