K0IN

agree, but do you think every dev you know konws this?

also "i just want to edit a config file and i want a nice ui", is how you get there.

a hile ago (when big providers still provided logprobs) i created a VS Code highlighter that visualizes unsure tokens.

Since most chat models want to answer with a human-readable message i think their logprobs are not as meaningful. It would be interesting to see if one choice is like "correct" and if the model wants to choose it more often, cause it might not answer the question but to prose to the user.

Does reinterpret_cast works in c++ for this?

I saw a lot of ppl think about what jev could use under the hood and could someone explain why this can't just be an embedding model where we just embed all the input + decisions and give back the cosine (or whatever) similarities?

My day job is improving data quality in big enterprises (we build products for this), and I see where this article is going, it's a bit watered down, but I see this every day at our customers, most companies just don't have a simple way to query their own data, create metrics, get ai to actually do stuff cause data is spread between 3-10 systems where 4 are legacy and don't even have an API and even if they can access they have like zero context 100 custom fields and suffer from no first time right inputs from the last 30+ years.

(ofc. I see only the ppl who need this kind of service so I'm biased, and maybe it's not a wide spread issue)

Good video on the topic: https://www.youtube.com/watch?v=gjHh9g7yo9Y

now add a source tab and lets see, how many ppl. will fix it themselves and how many turns it needs.

I have used the flash model for over 3b tokens and ofc. I saw some hallucinations and premature termination (I also get this on Astra - way more often than with deepseek v4 flash), but I never had a infinite loop (using the copilot as harness).

Am I missing something,I'm getting a 404?

upvote, k0ins hold together.

I just want to put this out there, smolmachines is a wonderful program to solve this, I use this mostly for stuff needing docker socket / docker in docker (example strix and agents). (I'm using podman on my host)

Me and my wife share calanders both ways between Google cal and iCloud cal its an absolutely shit show to setup, and I also rember having to use the Web app for certain settings how can we not have solved this by now?

what a nice platform, i would love this to patrole my house and have some fun.

first the automated part, that makes a pr "review ready", this helps us alot to gain confidence.

1. Automated tests that actually test real cases 2. Automated tests to test for regressions 3. Automated end-to-end tests, to make sure all core functions (and new functions still work and ui flows are stable / correct) 4. automated ai reviews (we use strix for security and a small Copilot SDK harness) 5. linters (for go: gosec,govuln) and formatters

For the human review process, we try to "never do big refactorings out of the blue", normally we do this in pair, or in tight loops. We dont have special reviewer roles, every dev can review any pr, we think, that everyone in a team is responsible for the product (and its quality, this also helps alot with knowledge transfare, we have retros and dojos to keep everyone on the same levle and align our teams "taste").

I was also startled when python did ß.upper() returns "SS". Which is kind of unsuspected in some cases (if string length changes with an upper call)

funny timing, i was exploring the same, some friends and me play alot of geogussr (a game where you put into streetview somewhere on the world, and have to find out where) lately.

I want to make it more video game like, cause the jumps of google maps make me motion sick.

I used google streetview as image source (cause you need to read signs, cars, see geometry) and colmap to reconstruct 3d chunks.

But it never really worked that well, jumps in time (cause the google car does not always continuously scan) and the image density is just too low, but was a fun side project.

I think this is our third lap.

you mean you dont want a singe company to have all your code?

yeah but using the rule of thumb (I see floating in the Internet) which is sqrt(#params * #active params) which would give sqrt(284B*13B) = 60B, so deepseek should perform better.

I used Qwen 3.6 27B extensively (>1B tokens) and DeepSeek V4 Flash (the older one also 2B+ tokens).

And I just can't fathom that the new 3.8 beats the new DeepSeek V4 Flash (which, in my eyes, is one of the best everyday coding models).

What an insane release, and convenient size to use every day/locally.

but i will test this model extensivly.

Some china proxies charge as little as 1 cent / Mio tokens for luna, which I use. As far as I know they bundle multiple codex subscriptions to get subsidized token pricings, so in codex it should be.

As other calculated, even with a Mi300 you could not saturae it enough with one stream to break even with the DS API, so I think renting sparks would make it even harder cause they are considerably slower.

(274gb/s vs 5.3tb/s)

I think it's nice to have options, but I also don't see the appeal in "sub x mb" (except as codegolf or just for the hacky part, for day to day this does not matter) especially if features are missing or other compromises have been made just for the sake of remaining small.

i use it too,can only recommend, also funny website btw.

Sick that you can get 2 arm cores and 12 GB ram for free at Oracle cloud, did not know that

Tldw: generating a general theme / color palette with gpt images -> using nano banana to split the images/generate single asset images -> creating 3d objects out of the images -> optimized the meshes using blender mcp

To make it lazy. It's the match anything in-between so you can put stuff before and after its my most used regex.

"my name is (.*?)$" => my name is k0in

Or values "last (.*?), was great" => last sunday, was great

Qwen 3.6 27b @ 90k context on a rtx 5090 (vllm)

So my favorite regex (.*?) works? Puh.

Is the classification model available, would be cool to have this in the toolbox.