carterschonwald

4 ms·

31.
by carterschonwald 3mo ago
5.6 sol is pretty good, its definitely very very well tuned, i'm not quite sure which of the others i should use near term, but theyre doing great work
32.
by carterschonwald 3mo ago
this so true. its really hard to make sure a model isnt going off the rails if i dont see full cot. the fact that oai and anthropic models hide it now has made them less reliable. which is a shame
33.
by carterschonwald 3mo ago
its a wrapper around mlx, so thats gonna be the portability bottleneck
34.
by carterschonwald 3mo ago
not sure about that, but im actively working on designing ultra sparse models that i want to have perform competitively with stuff 100-10_000 times larger. ehich does yield similar throughput. time will tell id it works out
35.
by carterschonwald 3mo ago
one of the most hilarious examples of how the current wh admins false protectionism, in my mind, is the crucible steel bankruptcy in jan/feb 2025. its a technological tragedy because it was the only facility im aware of globally that c
36.
by carterschonwald 3mo ago
ive had doom loops on release day with opus 4.6. quantization aint the culprit ;)
37.
by carterschonwald 3mo ago
even before the llm era sites would flag me as a bot for opening 15 links to read later. its fucking infuriating now
38.
by carterschonwald 3mo ago
this is pretty cool. i think part of the root cause is current rlhf post training design around confidence and optics rather than cooperative transparent honesty. though its kinda an expensive hypothesis to dig into as a private individual
39.
by carterschonwald 3mo ago
This paper points at an idea, but its really only legible if you have a more developed version of the idea already. I really should write more
40.
by carterschonwald 3mo ago
i literally had a co2 sensor for my engineering team last fall cause the space was so poorly ventilated. just measuring it continuously radically changed how everyone approached using the space packing wise and ventilation. smelled better
41.
by carterschonwald 3mo ago
good. Of course the precise language of the ruling matters, but good.
42.
by carterschonwald 3mo ago
so the most notoriously patent oriented tech firm is buying this up. lol ;) good for the founders. also explains why my resume got dropped on the floor as a desk reject :p
43.
by carterschonwald 4mo ago
.... i thought this was more widely known, granted i did write up a pretty wacky doc explaining way more fun experiments than these, and i have a fix that even prevents role collapse in my harness on github
44.
by carterschonwald 4mo ago
https://github.com/cartazio/oh-punkin-pi/blob/main/scripts/b... use that to install assuming you have whatevers needed to set up. ive a much fancier next gen thing i hope to make available as a sass
45.
by carterschonwald 4mo ago
“sit with” is way over used lately
46.
by carterschonwald 4mo ago
“worth sitting with ” is in the way to overused scale :(
47.
by carterschonwald 4mo ago
whats your fave/best one shot code gen? mine def has to be the initial impl of my python to my personal fave compiler ir. i got claude chat to write it in two sentences. 4 turns helping it rmeemver where it put shit becsuse of transc
48.
by carterschonwald 4mo ago
my heavily modified test bed for of oh my pi fixes this
49.
by carterschonwald 4mo ago
gemini models are also fantastic at understanding non spoken sounds
50.
by carterschonwald 4mo ago
this helps so much. i do it too. with some of the newer frontier models its unclear if you can even turn it off in the first party chat apps. havent compared api semantics yet.
51.
by carterschonwald 4mo ago
its part of making sure the model actually engages in emotive communication, if i'm inventing insults i've never even thought about, i'm furious :) saying i'm "furious" has lower entropy that incredibly implaus
52.
by carterschonwald 4mo ago
thats a harness issue not a model issue. eg i have my own reasoninf harness that forced persisted cot
53.
by carterschonwald 4mo ago
context window size isnt quite the issue though, its that the attention mass kinda spreads out too much and everything kinda converges to a sortah global average region full of what we know to be slop! theres some really cool ways at the ha
54.
by carterschonwald 4mo ago
i had the very strange experience last week of a recruiter listing your org as an example client, and when i looked stuff up i saw the current state.
55.
by carterschonwald 4mo ago
i actually think this is too tame. it really has to be stuff youd mever say to a real person.
56.
by carterschonwald 4mo ago
i want to view desktop on my mobile… and it wont lettt meeeee
57.
by carterschonwald 4mo ago
you might like the stuff in my work of oh my pi, its a test bed for my ideas around making these tools more reliable. hoping to maybe have a native ui iter of the real thing that this is a test bed for this summer. https://github
58.
by carterschonwald 4mo ago
the moment this clicked after reading it a few years ago its been my daily driver. that resource is a treasure
59.
by carterschonwald 4mo ago
i have some fun experiments i'm doing with full virtualization middle ware of all sys calls for agents tools/shell commands/io, still far from daily driver, but allows me to do a very rich overlay / virtual file system t
60.
by carterschonwald 4mo ago
I seem to have gotten an auto reject from your ATS before anyone technical could have read the application, which is odd because my career around building tools, libraries, compilers/interpreters plus associated experiments in applied

More ›