iandanforth

Deepseek-OCR uses SAM V1 as a component in its pipeline already. It also does layout detection.

I wonder if we'll get an updated DeepSeek-OCR that incorporates this. Would be very cool!

Not a fan. Code branches and this is Good Thing(TM). Result violates the single responsibility principle and tries to make what are distinct paths into a single thing. If your language has exceptions and returned values as distinct constructs then obfuscating them with Result means you end up fighting the language which becomes apparent fairly quickly. It's also a frustrating experience to want to interact with returned values directly and constantly have to deal with a polymorphic wrapper.

23 healthy adults ... who were able to fall asleep inside an MRI with EEG leads stuck to their heads. That's not an easy feat!

So now OpenAI is committed to spending $550 billion dollars? ($300B to Oracle and $250B to MS). If it currently has ~$10B in revenue / year, how on earth can it meet these commitments?

Isn't legal gambling default profitable? The house is allowed to remove players who are good at games that involve skill and set the win ratio on games that don't (as specified in regulations).

Is this a case of bureaucracy forcing people into illegality?

While all the security / privacy concerns in this thread are spot on I must say that this thing works. I had it do a proximity based search for me in Google Maps and then populate a new Google Sheet with this data and it just went off and did it right the first time. Now obviously Google could probably do this much better once they get around to it, but this is the first truly usable browser automation tool I've ever used and I spent years working with Selenium.

My plan is to create shadow accounts for Atlas and use it to automate tedious research tasks that span multiple websites that other AIs have trouble accessing.

Very cool, unfortunately I find the 3D completely unusable on mobile. The moment I touch it in orbit mode it locks to a southern pole view and whips about like crazy however I try rotate it.

I very much appreciate the style of this job post. Thank you!

I found and purchased a hand-made toy for my son this evening using ChatGPT searching Etsy. The search results were good and made it easy to pick from the many options. The checkout process using my stored CC hit an error (it didn't tell me what error) but when I switched to Google Pay it completed successfully. Overall a pretty nice experience as shopping goes.

I hate typing in Python. I spend a good chunk of my day fighting the type checker and adding meaningless assertions, casts, and new types all to satisfy what feels like an obsessive compulsive nitpicker. "Type partially unknown" haunts my dreams.

Duck typing is one of the best things about Python. It provides a developer experience second to none. Need to iterate over a collection of things? Great! Just do it! As long as it is an iterable (defined by methods, not by type) you can use it anywhere you want to. Want to create a data object that maps any hashable type to just about anything else? Dict has you covered! Put anything you want in there and don't worry about it.

If we ended up with a largely bug free production system then it might be worth it, but, just like other truly strongly typed languages, that doesn't happen, so I've sacrificed my developer experience for an unfulfilled promise.

If I wanted to use a strongly typed language I would, I don't, and the creeping infection of type enforcement into production codebases makes it hard to use the language I love professionally.

Lots of fun! Just the right amount of gameplay.

This is a great writeup! There was a period where reliable structured output was a significant differentiator and was the 'secret sauce' behind some companies success. A NL->SQL company I am familiar with comes to mind. Nice to see this both public and supported by a growing ecosystem of libraries.

One statement surprised me was that the author thinks "models over time will just be able to output JSON perfectly without the need for constraining over time."

I'm not sure how this conclusion was reached. "Perfectly" is a bar that probabilistic sampling cannot meet.

"Google said in a statement: “Quality raters are employed by our suppliers and are temporarily assigned to provide external feedback on our products. Their ratings are one of many aggregated data points that help us measure how well our systems are working, but do not directly impact our algorithms or models.” GlobalLogic declined to comment for this story." (emphasis mine)

How is this not a straight up lie? For this to be true they would have to throw away labeled training data.

Thanks for sharing that extensive documentation!

996 or worse is what got us the labour movement in the first place. In a well regulated industry the practice is illegal or carries explicit expectations of overtime, time off, or both.

For evil-aligned founders it's important to realize that exploiting workers is one of the best trod paths to success. If you can get away with it without causing a revolt you're almost certainly getting more for your money. So being up front with 996 is absolutely in your interests. Being hard-core, 10x, cracked (pick your generation's euphemism) is just good marketing. Be prepared to cut anyone who burns out, try to get people without any ties or those who can't afford to quit. Maybe even create a cult of personality around yourself.

This is McCarthyism. You take a polarizing word, then you attack your enemies by claiming they are that thing, and couch the whole thing in an "investigation" whose outcome is predetermined.

There is no merit to discussing if the target is that thing, it doesn't matter. It's an ideological attack. If you take it on its face then the attackers win because you're treating them as if they were honest participants in a discussion, which they are not.

And remember even if the investigation (which is a farce) goes nowhere, allowing it to exist unchallenged means that some people are going to be harassed and intimidated. But, that too is the point, fear is what they want.

Brian Armstrong is an unabashed ahole and here he's doing more ahole things. Nothing surprising.

I recognize that uv is very popular but it still feels faddish.

I don't want my package manager to format anything.

I don't want my package manager to lint anything.

I don't want my package manager to `run` anything.

I don't want my package manager to even manage environments (yes yes this one is a bit extreme).

I want to have single purpose tools that work together well.

This way I can have a flexible, composable toolchain that can both adapt to me and my specific project.

For example, I like conda, pip, ruff, mypy, flake8. Someone else might like venv, pip, black, pyright, flake8.

This is the heart of the Unix Philosophy and I feel like it's a philosophy each software community has to reject, suffer due to that rejection, and re-adopt on a regular basis. I'd rather we skip the suffering bit.

Any time you encounter a self help piece ask yourself, "Does this advice assume I have significant motivation, concentration, and persistence?" If so, drop it right in the trash. If you had that 9/10 you'd be doing something pretty effective already.

Another win for sheep! If you've never encountered the wonder substance known as lanolin let me bring you some good news. This other sheeps wool extract is an oil that is absurdly good at healing dry and irritated skin. It's also fantastic for hair, producing manageable and silky locks. I know I sound like a commercial, sorry! It's usually marketed as nipple cream, which I think has unfairly limited its other uses. If you're a parent you may know about it, otherwise try it yourself!

I tried the following prompt and other than producing a four panel comic that was black and white it completely ignored every other instruction. This was with 4 ultra. Maybe someone else will have better luck but the failure seemed stable.

''' A four panel comic strip. Simple black on white. Stick figures for characters. In the first panel there is a stick figure man and a stick figure bird eating bird seed at his feet. He is slightly hunched over to show he is looking at the bird. In the second panel. He is more hunched over looking more closely at the bird. In the third panel he is even more hunched over practically with his head to the bird, he is crouched down, knees bent, hands on thighs. In the upper left of the third panel the tip of an enormous beak can be seen, but it's only a few lines so could be anything. In the final panel the beak has gobbled up the man and his arms and legs are flailing outside of the beak while the small bird continues to eat birdseed on the ground. '''

My wife caught a glimpse of this over my shoulder, "What an ugly truck." she said immediately. Pretty much sums it up.

How is this kind of analogy helpful? You can frame any optimization problem as RL if you try hard enough. RL is a method of optimization which calls the optimum "reward maximization". You can craft the reward function any which way you want.

The key point about RL is that it is a sequential decision making process. If you don't have something (an agent) making multiple decisions over time while interacting with an environment, then why bother calling it RL?

The most interesting and divergent part of this post is this bit:

"Don’t use agents or things like editor with integrated coding agents."

He argues that the copy/paste back and forth with the web UI is essential for maintaining control and providing the correct context.

Points for being the first job post I've seen mentioning LeRobot!

"Every game needs a goal. In a Turing machine, goals are imposed from the outside. For true induction, the process itself should create its own goals. And there was a key constraint: Putnam realized that the dynamics he had in mind would only work mathematically if the system had just one goal governing all its behavior.

That’s when it hit him: The goal is to repeat. Repetition isn’t a goal that has to be programmed in from the outside; it’s baked into the very nature of things—to exist from one moment to the next is to repeat your existence. “This goal function,” Putnam wrote, “appears pre-encoded in the nature of being itself.”

So, here’s the game. The system starts out in a random mix of “on” and “off” states. Its goal is to repeat that state—to stay the same. But in each turn, a perturbation from the environment moves through the system, flipping states, and the system has to emit the right sequence of moves (by forming the right self-reinforcing loops) to alter the environment in such a way that it will perturb the system back to its original state."

I'm a big fan of this line of thinking. I've been arguing for years that RL should be based in homeostasis and this seems right along those lines. I wish I could have talked with him!

Man I saw the Fetch with its busted up gripper plastics and shocked to think they hadn't mothballed it and that it still worked!

Am I crazy or are the cats pictured under the graph swapped? The cat under the "leftward" bar is curled the right and the cat under the "rightward" bar is curled to the left. Did they just decide that egocentrism in directions doesn't apply to cats?

This is demonstrably false. Natural language acquisition is almost entirely listening and talking. The fastest and most consistently effective way to learn a language is immersion. The reason traditional material doesn't attempt immersive techniques is because it is much much easier to print a static book than it is to produce interactive and adaptive content.

The promise and potential of LLM based language learning apps is that you can cross that gap to full immersion in a way that has never been possible before.

Please be more ambitious.