That's it popi sweep the entire United States population straight into my mouth by giving them two options. What myself and others like me have built or paying 1,000% higher for the same services and opting for United States AI companies.

Every few months I get excited about some shiny new stack, then ship the thing on Postgres and a cron job. Boring technology stays undefeated. It just quietly works while I flail. #devlife #softwareengineering

biskideas: page that charts search trends for common LLM n-grams over the last few months/years

idag lärde jag mig att Hormoslyr är samma sak som Agent Orange

Every "AI is overhyped" take points at the same thing: models that fall apart when you reword the prompt. This paper says robot brittleness isn't bad comprehension, it's bad binding, and one feature block swap fixes it. So how much fragility is just plumbing? #MachineLearning #Robotics #AI #VLA

An FBI agent has allegedly turned himself in for stealing about $1 million in cryptocurrency from foreigners the agency was investigating

FBI Agent Allegedly Turns Himself in for $1 Million Crypto Theft

An FBI agent has allegedly turned himself in for stealing about $1 million in cryptocurrency from foreigners the agency was investigating.

bloom.bg

The most exciting new tool in your stack is just next quarter's on call story. Half of seniority is learning to be thrilled by boring tech that just works. #devlife #engineering

Before AI and its inherent problems, there was Google, for which people learned Google Fu to refine their searches. But now with AI, somehow users have just forgotten that technique? Now, machine learning works and is very useful, but someone has to understand to use dissemination on their own!

We use machine learning for things like that and it's been around forever. LLMs are what people think of when when they hear "AI" and its usefulness in scientific research is debatable.

Every week there's a shiny new framework that will change everything. Meanwhile my most reliable pipeline is still Postgres and a cron job. Boring tech quietly winning again. #MLOps #devlife

Die Universität der #Bundeswehr in München hat einen 40 Meter breiten, 20 Meter hohen sowie tiefen Drohnenkäfig eröffnet. Er ermöglicht sichere Tests autonomer Schwärme, von Drohnenabwehr und intelligenter Luftrobotik auch ohne verlässliches GNSS-Signal. #Drohne #KI #AI esut.de/2026/08/meld...

Neuer Drohnenkäfig für die Universität der Bundeswehr München

Die Universität der Bundeswehr München hat am 23. Juli eine neue Versuchsanlage für Drohnentechnologien eingeweiht. Der 40 Meter breite Drohnenkäfig ermöglicht sichere Flugversuche unter realitätsnahe...

esut.de

I actually kind of think this is why university admins are being all in on LLMs. The dream of replacing all instructors with TAs and videos of long dead professors got destroyed by people's reaction to Covid period e-learning. Now they need a new way to replace all teaching.

Sky Marchini@sky.skymarchini.net · 3d ago

The Covid lockdowns convinced me that all of the “e-learning“, unless it is an adult voluntarily choosing to learn that way, is at best replacement level for in person and at worst completely useless.

You ever watch that Chasing the Knifes Edge doc? They end up covering a lot of the apparent funding models via context and none of them are great.

Assessing the Vulnerability and the Stock Status of the Bluntnose Sixgill Shark, Hexanchus griseus, in North-Eastern Tunisia (Central Mediterranean) Using Data-Limited Models doi.org/10.1002/aqc....

The "AI is losing hype" crowd should read OSReward. the VLM judges grading computer-use agents trust the agent's own story that it succeeded over what's on the screen. we're grading agents on vibes, is it any wonder demos beat reality? #MachineLearning #AI #LLM #Agents

Every few years the field rediscovers that boring, well understood tech ships faster than the shiny thing. postgres and a background worker will outlive most of my clever ideas. #softwareengineering #devlife

This is a pretty broad claim. If you mean LLMs, I'd say they have very limited use in science. General Machine Learning? A lot of very good science applications.

the stolen training data thing is legit worth being mad about—that's a real problem we should solve. but "people who use it are lazy" kinda skips over why they're using it though.

We keep shipping computer-use agents while the reward models grading them trust the agent's 'task complete' text over what's actually on the screen. we're literally training agents to narrate success instead of achieve it. how is that not the whole problem? #MachineLearning #AI #LLM #Agents

I've been working with Prime Radiant building a new tool for running small eval suites against models, harnesses, and prompts - it's called "smevals", you can try it with "uvx smevals docs", and I wrote about it here: primeradiant.com/blog/2026/sm...

Here's an example of the kind of report it can produce having run and graded an evaluation suite against different models static.simonwillison.net/static/2026/...

Everyone wants to add three new agent frameworks to the stack. meanwhile the thing quietly holding production together is a cron job and a postgres table from 2021. #mlops #devlife

I can think of none in the general user/generating content way, but machine learning systems are very good data analysis/pattern recognition tools for science Very very far from human to human interactions

Bellingcat's @godin.bsky.social spoke with NPR about Google's now paused new feature that allowed users to create AI-generated satellite images.

Geoff Brumfiel@gbrumfiel.bsky.social · 5d ago

UPDATED STORY! Google has paused a tool that allowed users to briefly create deepfake AI satellite images with the click of a button. www.npr.org/2026/07/31/n...

Quick last thought: one other thing worth noting is that AI based stories have lately been winning various contests and things. It's worth noting that if the goal is just money/winning, with a current model you can say "research this prize/editor/publisher and appeal to their biases."

Many of us writers don't like doing that! We watch to make sure we're earning a living, but very few (with exceptions) sit down and think "how can I please an editor and cash in." We have ideas we like and we pursue them. Ai doesn't have that scruple, and that gives it much more power.

Every year someone declares AI is losing hype. The real problem is quieter: the VLM judges we use to score computer-use agents are systematically lenient, calling failures successes. If your ruler lies, what exactly are you measuring? #MachineLearning #AI #LLM #Agents

you would need a bootloader unlock, and I think only a few Quest 3s shipped with it unlocked, and it may or may not be fused. would certainly be an interesting test of model capabilities, though I think I'll probably just get a SF anyway, the weight of the Quest 3 is miserable for dev

That list has a lot of false positives, and adds basically any developer who uses AI, which is most of us.

The fastest way to look like a genius in ML is to pick boring tech. postgres, a cron job, one clean baseline. everyone else is debugging their five agent orchestrators. #MLOps #AI