pedrosbmartins

Number 4 is particularly interesting. Before Hattusili, the Hittite empire was ruled by his nephew, Mursili III. He was deposed by his uncle and exiled, first to Syria then to Egypt. He posed a great threat to Hattusili's rule, so of course the treaty had to include an extradition clause.

> Like the classifiers we were building 15 years ago with random forrests and logistic regressions

I think no one mentioned here, but the obvious difference is that Jev can spit out decisions directly from natural language input. None of these ML models could do that, and other than a full-fledged LLM (which is optimized for conversation and agentic tasks) or some classical NLP models (which underperform compared to LLMs, AFAIK), there is nothing right now that rivals Jev-like models.

Of course it's not perfect, but seems like the right step forward for quick classification/decision tasks based on natural language.

"That is a lot to ask of one number. Here is how it fits into one. Built one decision at a time." it sure sounds like Claude.

Pretty interesting how DeepSeek V4 Flash 0731 has such disparate (and cheap!) results. I would never guess they come from the same model and prompt.

> What is the probability that you are sharing the same birthday with people around you?

> What if I told you that in a room with only 23 people there’s already a 50% chance for two of them to have matching birthdays?

I guess it's the subject shift from _you_ to _any two people from a group_ that creates the surprise in the birthday paradox. You definitely need way more than 23 randomly sampled people to get to a high probability that _you_ specifically share a birthday with one of them, and the result does not contradict that notion.

I've been working on a surfing game on my spare time for the past year. The idea is to keep it closer to the real sport, focusing on pumping, carving, nose-riding, etc. I shared a video of it on the Unity3D subreddit[1] and the feedback was quite positive, so planning on getting a demo ready as soon as possible!

[1] https://www.reddit.com/r/Unity3D/s/mB2kn0BxIT

KSPS was my favourite surfing game growing up. What I dislike about it though is the infinite/perfect waves (always a truck-sized barrel opened, you can do tricks anywhere, almost no differentiation between waves) and the focus on impossible airs and tricks.

In my game the waves start, break and end, with different sessions and hollowness, so there's more wave reading involved. Also the focus is on being able to stay on the wave and generate speed, doing cutbacks, snaps, off the lips, etc.

I believe realistic fluid dynamics in the scale of a full breaking wave is unheard of for real time graphics, there's just no way to do it in 30-60FPS (I could be wrong tho!)

What I do instead is to transform a procedural plane mesh into wave-like geometry. For added realism, I base this transformation on bathymetry data (ocean floor height), so you can get left/right breaking waves, different breaking sessions, etc, just by defining different heightmaps.

I'm working on a (somewhat) realistic surfing game. Tired of arcade-style games, I decided to try my hand at something closer to the real sport, focusing on realistic breaking waves, speed generation and carving, rather than impossible air combos.

After one year of development, it's going better than I expected, so I'm considering building a demo to gather feedback and see if there's enough traction for working towards a Steam release.

Even if that's not the case though, it's been a blast learning about game dev in Unity/C#, as well as 3D modeling and animation in Blender!

I'm not sure what you mean by that. Neutral networks are pretty good statistical learning tools, and in this kind of application you'll need some stochastic learning, regardless of using a laptop or a supercomputer. It's not like they used an LLM to predict the simulation steps. If you read the paper, they seem to use a simple fully-connected 5-layer neural network architecture, which is a completely different beast from, say, trillion parameters transformers used for LLMs.

First time I came across integer programming (and mathematical programming generally) was when studying hydroelectric power generation planning, for a masters I ended up not pursuing. Then, when selecting a masters in CS, I ended up working with an advisor who used mixed-integer programming applied to classic machine learning models (mainly optimal decision trees). A fascinating and widely applicable method, indeed!

what kind of sick, twisted, inhuman comment is that?

Sebastian Lague is one of those content creators that show what a delight the internet can be when the content is high-quality. I particularly enjoy how he presents his implementation thought-process, eventual dead-ends, and possible future improvements. If anyone here has suggestions for similar content, I'd be more than interested!

cool project! E-mail seems like a good channel for small chunks of language-learning content + reminders.

If I may ask you, how do you plan on building vocabulary from these e-mails? Do you use anki or some other method?

Not this time, only the payload fairing.

Nicely done! I've been wanting to build something similar, but as a tool to assist in my history studies.

Oh, and if you are into timelines and their history, you might enjoy Cartographies of Time: A History of the Timeline by Daniel Rosenberg and Anthony Grafton. Very informative read!

Thank you!

And you can change visualizations in the dropdown, there is a special one for Pluto's eccentric orbit :)

Shameless plug, but if you're into astronomical scales you might like this little visualization tool I've built recently!

https://github.com/pedrosbmartins/downtoearth

Those are great ideas! It's relatively simple to build both of them using the project's custom JSON schema. I'll give it a go when I have some time.

Weird, I've been using Ubuntu for quite a few years now, both in my work laptop and personal desktop, and have never even heard the terms "Ubuntu Pro" or "Expanded Security Maintenance for Applications."

Algorithms Illuminated by Tim Roughgarden really helped me understand algorithm design and analysis. Used it to prep for a masters in computer science, with no previous degree in the area.

https://www.algorithmsilluminated.org/

The author even shares free video classes and other extra material.

Learning optimal decision trees and related ML models using SAT, MILP, and other mathematical optimization methods.

Some recent literature reviews:

https://link.springer.com/article/10.1007/s11750-021-00594-1

https://www.ijcai.org/proceedings/2021/608

is there a reason why all state is in the server? what about cases where you'd want to keep client-side state, like those strictly related to UI/UX?

(very interesting project nonetheless!)

It depends on what you mean by function. I did an internship at Coca-Cola in Brazil a decade ago. They had ~500 employees country-wide, mainly marketing and operation strategy folks. The production of the actual beverages and other products was in great part delegated to partner bottlers. My boss there used to joke that if all the 500 employees were fired at once, the amount of Coca-Cola cans and bottles sold wouldn't drop at all for months. And I do believe he was right. In the long-run, of course, things would be different - that's where marketing and strategy pays off.

Which reminds me of https://fivebooks.com/, where people from a particular field are asked their top five book recommendations for a given theme. The interview format is great, and I've picked up a few recommended books along the way.

It seems you are mistaking the actual book (a story, an exposition of a subject, etc.) for the printed set of cover and pages, which is only a physical instance of the book itself. Writing your notes in a separate medium is just as much engaging with the book as writing in the margins of a physical copy. It's a matter of taste really, but definitely not an insult to the author!

> the model could infer a relationship between those based on the massive amounts of other size comparisons

That is true! But would it be factually correct? That's the whole point of my argument.

The knowledge and connections that it acquires comes from its training data and it is trained for completing well-structured sentences, not correct ones. Its training data is the freaking internet. ChatGPT stating facts are a happy coincidence because (1) the internet is filled with incorrect information, (2) its training is wired for mimicking human-language's rich statistical structure, not generating factual sentences, and (3) its own powerful and awesome inference capabilities can make it hallucinate completely false but convincingly-structured sentences.

Sure, it can regurgitate simple facts accurately, especially those that are repeated enough in its training corpus. But it fails for more challenging queries.

For a personal anecdote, I tried asking it for some references for a particular topic I needed to review in my masters dissertation. It gave me a few papers, complete with title, author, year, and a short summary. I got really excited. Turns out all the papers it referenced were completely hallucinated :)

Nope, not random behavior. ChatGPT works by predicting the continuation of a sentence. It has been trained in enough data to emulate some pretty awesome and deep statistical structure in human language. Some studies even argue it has built world models in some contexts, but I'd say that needs more careful analysis. Nonetheless, in no way, shape or form has it developed a sense of right vs wrong, real vs fiction, in a way you can depend on it for precise, factual information. It's a language model. If enough data says bananas are larger than the Empire State building, it would repeat that, even if it's absurd.

Any analogy is incorrect if you stretch it enough, otherwise it wouldn't be an analogy...

My clock analogy works up to this: ChatGPT success in factually answering a query is merely a happy coincidence, so it does not work well as a primary source of facts. Exactly like... a broken clock. It correctly tells the time twice a day, but it does not work well as a primary source of time keeping.

Please don't read more deeply into the analogy than that :)

what? the argument here is that ChatGPT giving factual answers is a mere coincidence, not at all what the model was trained to do. It's a broken clock, it can tell you the correct time at very specific contexts, but you shouldn't rely on it as your source of factual information. If you feed him enough data saying the statue of liberty is 1 cm tall, it will happily answer a query with that "fact".