endymi0n

law-bearing

Been using a lot of Kimi K3 lately and the answers have been… „load-bearing“ to the point of hilariousness. It‘s obvious from where they distilled, even if sceptics rightly point out it can‘t have been the only source of their secret sauce, as it‘s been better than the current Opus 4.x at the time of release.

Having used Kimi K3 intensively for the past week, I can vouch for load-bearingly close.

Finally. I‘d be delighted though if they actually implemented language autodetection (like everywhere else) though. There’s little more frustrating in my day to day than having dictated half a page to find that it‘s complete gibberish because Apple forces you to select the right language first…

Now that's a very high ratio of personal beliefs to character ratio. I'd be delighted to see some evidence behind your opinions.

The whole use of the word “use” throws me off. It’s not like the water just disappears. It’s still very much there, just… well, yep, what exactly? Dirtier? Evaporated? Warmer? We’re drinking water every day from the tap that has previously been “used” as fish pee, nuclear plant cooling water and sawmill fuel. I’m not too dead yet and I think it would be great to get a more scientific discussion from public media.

not parent, but I kinda have the same thoughts often. Maybe I can’t do inference on them in the same form factor (yet!), but just the fact that the weights of a model that comes close to capturing a close enough approximation of the combined knowledge, experience and intelligence of mankind fits onto a MicroSD never fails to amaze me.

Kids and their wellbeing and education might be a good start…

there’s a lot of open models out there… I told Claude to do a weighted score on several models and deduplicate by CLIP similarity for an expedition, should be easy to replicate (see below). Sure doesn’t select the absolute best pics from an emotional impact perspective, but it was pretty damn good at me not having to wade through the bottom 80% of mediocre shots and dupes!

—-

“Models scored all 4,487 photos. NIMA rewards technical craft (sharpness, composition), LAION rewards emotional/aesthetic appeal, MUSIQ is more general quality. Combined: 0.4 NIMA + 0.3 LAION + 0.3 MUSIQ, deduped at 0.85 CLIP similarity.

Interesting: the models wildly disagreed on some shots — one photo ranked NIMA #2 globally but LAION #4313.”

To paraphrase Gwynne Shotwell: “Not too bad for just a large Markov chain, eh?”

I don’t exactly know where MTP inference fits within the inference stack, but does someone know whether it’s possible to implement it for the MLX universe?

OpenAI is the first company that has reached a level of intelligence so high, the model has finally become smart enough to make YOU do all the work. Emergent behavior in action.

All earnesty aside, OpenAI’s oddly specific singular focus on “intelligence per token” (also in the benchmarks) that literally noone else pushes so hard eerily reminds me of Apple’s Macbook anorexia era pre-M1. One metric to chase at the cost of literally anything else. GPT-5.3+ are some of the smartest models out there and could be a pleasure to work with, if they weren’t lazy bastards to the point of being completely infuriating.

I had this funny moment when I realized we went full circle...

"INTERCAL has many other features designed to make it even more aesthetically unpleasing to the programmer: it uses statements such as "READ OUT", "IGNORE", "FORGET", and modifiers such as "PLEASE". This last keyword provides two reasons for the program's rejection by the compiler: if "PLEASE" does not appear often enough, the program is considered insufficiently polite, and the error message says this; if it appears too often, the program could be rejected as excessively polite. Although this feature existed in the original INTERCAL compiler, it was undocumented.[7]"

— https://en.wikipedia.org/wiki/INTERCAL

I’ve had great success replacing it with Kimi 2.6

well, I do understand the core motivation, but if the system prompt literally says “I am not budget constrained. Spend tokens liberally, think hardest, be proactive, never be lazy.” and I’m on an open pay-per-token plan on the API, that’s not what I consider optimal behavior, even in a business sense.

Did you guys do anything about GPT‘s motivation? I tried to use GPT-5.4 API (at xhigh) for my OpenClaw after the Anthropic Oauthgate, but I just couldn‘t drag it to do its job. I had the most hilarious dialogues along the lines of „You stopped, X would have been next.“ - „Yeah, I‘m sorry, I failed. I should have done X next.“ - „Well, how about you just do it?“ - „Yep, I really should have done it now.“ - “Do X, right now, this is an instruction.” - “I didn’t. You’re right, I have failed you. There’s no apology for that.”

I literally wasn’t able to convince the model to WORK, on a quick, safe and benign subtask that later GLM, Kimi and Minimax succeeded on without issues. Had to kick OpenAI immediately unfortunately.

at this trajectory, unsloth are going to release the models BEFORE the model drop within the next weeks...

Welcome to npm post-install scripts... https://docs.npmjs.com/cli/v11/using-npm/scripts

I've come to dread any formalization of Agile. Agile development is fine. I've built a 40+ engineering team with it. I can vouch for its effectiveness when applied to small, excellent teams.

For reference, here's all the Agile you need, it's 4 sentences:

Individuals and interactions over processes and tools

Working software over comprehensive documentation

Customer collaboration over contract negotiation

Responding to change over following a plan

The real problem is that capital-A Agile is not agile at all, but exactly the opposite: A fat process that enforces following a plan (regular, rigid meeting structure), creating comprehensive documentation (user stories, specs, mocks, task board) and contract negotiation (estimation meetings, planning poker). It's a bastardization of the original idea, born by process first people who tried to copy the methods of successful teams without understanding them.

I've experimented quite a bit with mem0 (which is similar in design) for my OpenClaw and stopped using it very soon. My impression is that "facts" are an incredibly dull and far too rigid tool for any actual job at hand and for me were a step back instead of forward in daily use. In the end, the extracted "facts database" was a complete mess of largely incomplete, invalid, inefficient and unhelpful sentences that didn't help any of my conversations, and after the third injected wrong fact I went back to QMD and prose / summarization. Sometimes it's slightly worse at updating stuck facts, but I'll take a 1000% better big picture and usefulness over working with "facts".

The failure modes were multiple: - Facts rarely exist in a vacuum but have lots of subtlety - Inferring facts from conversation has a gazillion failure modes, especially irony and sarcasm lead to hilarious outcomes (joking about a sixpack with a fat buddy -> "XYZ is interested in achieving an athletic form"), but even things as simple as extracting a concrete date too often go wrong - Facts are almost never as binary as they seem. "ABC has the flights booked for the Paris trip". Now I decided afterwards to continue to New York to visit a friend instead of going home and completely stumped the agent.

It's definitely a bit ironic that a war for oil drives the last push for getting rid of it, but I'll take that as well, if logic and sanity didn't help ¯\_(ツ)_/¯

That benchmark is as outdated as completely unrealistic, as if invented by the oil/nuclear industry. Obviously 100% pure solar generation will be completely unfeasable in a place as Germany, but that completely misses the point that a realistic combo of solar/wind/biomass has a FAR higher combined capacity factor than solar alone.

Also, it's based on 2021 (or before) storage cost figures, which have halved in the meantime. https://assets.bbhub.io/professional/sites/44/LCOE-11.png

I call BS.

Nuclear fans are heavily underestimating the cost of that energy source. The levelized cost of energy per kWh today is TRIPLE that of solar already, at a negative learning curve, with a gap only widening (and accelerating so). For the very same cost per kWh, you can get double overprovisioned solar PLUS battery storage at 90% capacity factor, TODAY.

All of that fully decentralized, within the next years instead of decades, with distributed (not megacorp) ownership AND not having every other of these megaprojects cancelled due to protests.

And that figure doesn't even include externalized cost like national/environmental security or decommissioning costs.

Nuclear is riding a dead horse in 2026.

"don't ever lie about your past compensation" — because they can't figure it out on their own and IF they do (at least in my jurisdiction), you've got a nice case on your hands to sue them for violating privacy laws.

The correct answer is: ALWAYS lie about your past compensation. It's the only way to get forward, one way or the other.

Well, let's not forget the conflict of interest on the other side as well, of someone having invested decades of professional experience into a very lucrative field already getting obliterated by AI in some narrow fields.

Getting rid of radiologists is as much nonsense and saber rattling as suggesting using AI would harm patients.

The answer is clearly just the same as in software development or any other AI impacted field: Let the best professionals handle 10x+ the volume. What that means for all the rest of employees is the question of the century though...

„Open Access APIs are like a subway. You use them to capture a market and then you get out.“

— Erdogan, probably.

G is tame. Wait until you hear of Databricks’ Series K…

https://www.thesaasnews.com/news/databricks-raises-1b-series...

One. Trillion. Even on native int4 that’s… half a terabyte of vram?!

Technical awe at this marvel aside that cracks the 50th percentile of HLE, the snarky part of me says there’s only half the danger in giving something away nobody can run at home anyway…

I hate NOTHING quite the way how Claude jovially and endlessly raves about the 9/10 tasks it "succeeded" at after making them up, while conveniently forgetting to mention it completely and utterly failed at the main task I asked it to do.

Meanwhile, core functionality like “Find My” is completely and utterly broken. Leaving behind my stuff at a new place gets me at least two different messages on my Apple Watch at different timing. One for my devices, one for my Apple tags. One only has a “dismiss” button, the other has a “trust location” button that when I click, it says “content unavailable”, and if it works (which is only! over Wi-Fi), then it only works for that one device. I always need to go through the find my app at every new place since it’s an absolute UX disaster.

That’s what I get for carrying only Apple gear in the thousands of euros with me.