mitchellpkt

8 ms·

31.
by mitchellpkt 3y ago
Just played a round, pretty fun! Thanks for sharing.
32.
by mitchellpkt 3y ago
Update: I wrote a toy python library for evaluating Langton's ant in procedurally-generated infinite adversarial environments. https://github.com/Mitchellpkt/infinite_adversarial_langtons...
33.
by mitchellpkt 3y ago
I second this :) Lived in Madison in my 20's and loved it. Great people, great environment, great food. Easy to get around on foot, on bike, or by bus.
34.
by mitchellpkt 3y ago
I have been using chatGPT with GPT4 since it came out, and I have not experienced decreased quality. I wonder why you can’t get it to give approximations. I have not encountered difficulties there. I asked it to approximate the weight of ar
35.
by mitchellpkt 3y ago
No idea. Hmm, off the cuff maybe I’d conjecture that there cannot exist any initial configurations such that the ant gets stuck in a loop of finite length for infinite iterations. On any Nth iteration, the ant has only touched <= N tiles
36.
by mitchellpkt 3y ago
Yep, it's a deterministic perfect information game. To "win" one needs only produce an initial board (or board generating procedure) that achieves certain characteristics, such as bounded highway length even with infinite ite
37.
by mitchellpkt 3y ago
I keep wondering about the dynamics of Langton's ant within an infinite adversarial environment. The game I have in mind retains the standard behavior and movement patterns of Langton's ant. However, it introduces an adversarial c
38.
by mitchellpkt 3y ago
I haven't had a chance to test it personally, but I think this is a main premise of GitHub's Copilot CLI product. Quoting below from https://githubnext.com/projects/copilot-cli > Copilot for CLI provides t
39.
by mitchellpkt 3y ago
Keep an eye out for emails from noreply@githubnext.com On 27-Mar 2023 I registered for some of the "X" waitlists. On 13-Apr 2023 I received an email with the subject line "Welcome to the Copilot for CLI technical preview!&quo
40.
by mitchellpkt 3y ago
+1. To expand on your friction comment: I’ve been doing unique @myname.com for years and learned quickly to avoid putting the full company name to avoid confusion. Now, I just use the first 2 and last 2 letters, so for example hackernews wo
41.
by mitchellpkt 4y ago
There's a nice overview at [0] "An Analysis of Uniswap markets." Some very interesting larger dynamics are discussed in [1] "When Does The Tail Wag The Dog? Curvature and Market Making." The Uniswap mechanism varies
42.
by mitchellpkt 4y ago
There are still some performance-sensitive niches that you might enjoy. Large scale numeric simulation workloads can have a variety of bottlenecks, and the difference between the easiest solution and the fastest solution can be orders of ma
43.
by mitchellpkt 4y ago
Congrats on graduating, and good move building experience with the latest dev tooling like ChatGPT. I don’t think you should feel too discouraged. Plenty of companies do not allow their developers to use third party language models/ser
44.
by mitchellpkt 4y ago
I’ve been using pycharm pretty much daily for a few years, and switched to the new UI a few weeks ago. I like it, especially when doing dev work with only one monitor. You might hate it, but it’s worth giving it a shot.
45.
by mitchellpkt 4y ago
[No spoilers] This was a bit of a mind bender for me: I was halfway through "Revelation Space" by Alastair Reynolds when ChatGPT was released. Throughout the book, it is common for characters to use natural language to delegate te
46.
by mitchellpkt 4y ago
Arguably this is the case now. I don’t think there’s any way to access chatGPT-4 unless you have the financial bandwidth to allocate an extra $20/month for an OpenAI subscription.
47.
by mitchellpkt 4y ago
Sometimes it isn’t a big deal, depends on the context and whether unexpected failure modes have dire consequences. A black box AI algorithm that generates shirts in a video game? Sure, sign me up. Worst case scenario the shirt looks weird o
48.
by mitchellpkt 4y ago
Session messenger is a fork of Signal that ditches phone numbers for IDs and verification. https://github.com/oxen-io/session-android
49.
by mitchellpkt 4y ago
Anecdotally: I do a decent amount of DS work and use GPT to help me move faster through the mundane data tooling work like ingesting new file formats or updating schemas. This enables me to spend more of my time and energy on the interestin
50.
by mitchellpkt 4y ago
Just read the blog post, very interesting. I thought sections about changing the temperature and repetition penalty were particularly neat, thanks for including examples.
51.
by mitchellpkt 4y ago
It warrants professionally-cautious skepticism because the synthetic data generation process often involves assumptions, approximations, and spurious or omitted dynamics. These limitations can then impact the model, as it is learning on dyn
52.
by mitchellpkt 4y ago
Portal (2007) is timeless. Chess (600) has also held up pretty well.
53.
by mitchellpkt 4y ago
Grad school was the first time I had to get under the hood with numeric simulation code. Experience: It was a bit of a learning curve, but a manageable one. I think maybe 80% of my energy went into improving my domain expertise (physical ch
54.
by mitchellpkt 4y ago
I had a vivid imagination as a kid, lost it in my 20s, and regained it in my 30s. After decades of schooling, reading had beed reduced to a mechanical necessity. It took some work and intention to slow down and learn to actually imagine the
55.
by mitchellpkt 4y ago
A while back when I was familiar with coding but wanted to improve my python skills, I found this video helpful: https://www.youtube.com/watch?v=qUeud6DvOWI Once I understood fstrings, list/dict/set comprehension,
56.
by mitchellpkt 4y ago
One of my previous jobs involved tracking a lot of threads across a lot of emails. It would have been handy if we had a private LLM that was trained/prompted to classify emails according to internal systems, like custom CRM stages. I w
57.
by mitchellpkt 4y ago
I use a number of GPT-based products on a day-to-day basis and find them incredibly useful, so unfortunately I can't assuage your concerns by downplaying current and future language model technology. The great news is that many of thes
58.
by mitchellpkt 4y ago
I don’t mind ads that are in the margins or added to a list of results that I’m scrolling through anyways. These don’t feel intrusive because they don’t break the flow I would have without ads. On the other hand I become frustrated when ads
59.
by mitchellpkt 4y ago
I agree. There are various complexity downsides (implementation, key management, etc) but it should be fine from a practical cryptographic security perspective. While I can't offer you a formal cryptanalysis, I do have a handwave-y tho
60.
by mitchellpkt 4y ago
Good questions. Hmm do you think maybe it would be viable if the payouts were based on [model] performance rather than ostensible training time? There's a useful asymmetry we can exploit: finding weights that perform well is computatio

More ›