richdougherty

6 ms·

1.
by richdougherty 11d ago
You probably need a programming language more powerful than Rust that can encode and check the complicated invariants needed by the JIT. There are classes of type system—"dependently-typed"—where you can encode arbitrary predicate
2.
by richdougherty 7mo ago
If you'll excuse the self-post, here's a blog post on goto with delimited continuations. https://rd.nz/2009/03/goto-in-scala.html It uses an experimental compiler plugin for the Scala compiler. It's
3.
by richdougherty 1y ago
Seems likely because the article says "Meta’s deal with Google is mainly around artificial intelligence infrastructure, said one of the people".
4.
by richdougherty 2y ago
"The cost-effective nature of AI makes it highly plausible we're moving towards an agent vs agent future." Sounds right. I assume we will all have AI agents triaging our emails trying to protect us. Maybe we will need AI to h
5.
by richdougherty 2y ago
Kudos to the dev for coming up with the eye position fixing solution. Building further on this idea, I wonder if instead of changing the image to look at the camera, we could change the "camera" to be where we're looking. In
6.
by richdougherty 2y ago
So it's also vulnerable to a Helicopter Injection Attack?
7.
by richdougherty 2y ago
If you have control of the tokenizer you could make sure it doesn't produce these tokens on user input. I.e. instead of the special "<eos>" token, produce something like "<", "eos", ">&quo
8.
by richdougherty 2y ago
Source with more information: https://www.tuxedocomputers.com/en/TUXEDO-on-ARM-is-coming.t...
9.
by richdougherty 2y ago
A few details here: "Recall leverages your personal semantic index, built and stored entirely on your device. Your snapshots are yours; they stay locally on your PC. You can delete individual snapshots, adjust and delete ranges of time
10.
by richdougherty 2y ago
I can't track down the citation (either Google or DeepMind I think), but I remember reading research from a year or two ago how adding extra languages (French, German) improved English language performance. There may have also been an
11.
by richdougherty 2y ago
Related? "Let's Think Dot by Dot: Hidden Computation in Transformer Language Models" https://arxiv.org/abs/2404.15758 > Chain-of-thought responses from language models improve performance across
12.
by richdougherty 2y ago
Anyone who enjoyed this, might also like "All the Horses of Iceland". It's a fictional work, but written by a medieval scholar. It's set in the 9th century where a Norse trader travels through an alternative central Asia
13.
by richdougherty 3y ago
http://ymeskhout.substack.com/p/eleven-magic-words-unlocked
14.
by richdougherty 3y ago
Agree, nesting files with cat >Dockerfile <<'EOF' and having a basic bash script seems way nicer than putting all the shell logic on the #! line.
15.
by richdougherty 3y ago
https://twitter.com/ylecun/status/1728126868342145481 "Please ignore the deluge of complete nonsense about Q . One of the main challenges to improve LLM reliability is to replace Auto-Regressive token predict
16.
by richdougherty 3y ago
Site not loading for me, but probably this: https://nothing.tech/pages/nothing-chats
17.
by richdougherty 3y ago
Just speculating, but perhaps share prices would drop until the returns per dollar invested recovers to the desired level. So it might be like a one off shock to investors who hold shares now, since the new equilibrium has shares priced low
18.
by richdougherty 3y ago
Supported in Java 10 (and backported to Java 8) since 2018. Not sure about .NET. - "The JVM has been modified to be aware that it is running in a Docker container and will extract container specific configuration information instead of
19.
by richdougherty 3y ago
https://math.stackexchange.com/users/97378/cleo
20.
by richdougherty 3y ago
https://en.m.wikipedia.org/wiki/Generalized_second-price_auc...
21.
by richdougherty 3y ago
Agreed, it needs to be Android based - the other option would be to implement iOS APIs, but Android would be a lot easier. An analogy would be if you make a browser, to use standard HTML and HTTP, not try to invent your own thing.
22.
by richdougherty 3y ago
https://arxiv.org/abs/2310.02207
23.
by richdougherty 3y ago
A really interesting article! https://thegradient.pub/othello/ "How do these models achieve this kind of performance? Do they merely memorize training data and reread it out loud, or are they picking up the rules
24.
by richdougherty 3y ago
I'd love them to fund an Android variant or replacement for Android services. Mobile is the other main platform apart from web.
25.
by richdougherty 3y ago
You can run them either for general purpose inference. You can also fine-tune them and get improved performance for specific use cases. It's safe to assume they're worse at every task than larger models, so I wouldn't look at
26.
by richdougherty 3y ago
I haven't read the article, but "The Winner's Curse" is an economics concept I'm familiar with. Basically, The Big Insight is that every bidder has an estimate of the value of something, and the true value might dif
27.
by richdougherty 3y ago
Yes, no one's saying that people evolved a predisposition for participating in genetic studies specifically. It'll be some other traits like you suggested. The research is just saying "these databases are not randomly selecte
28.
by richdougherty 3y ago
I guess that's why the article says at the end "The primary barrier to entry for farmers is a 500,000 rupiah ($33) weeding machine."
29.
by richdougherty 3y ago
Which links to the original Phoronix source. Conclusion from Phoronix: "Overall it comes down to what workloads you are engaged in whether you may notice any performance difference when upgrading your Linux kernel (or otherwise being p
30.
by richdougherty 3y ago
Another interesting C++ successor language: https://github.com/carbon-language/carbon-lang

More ›