Dynomight

It is an interesting article and some of the results are unexpected, but the layout is way too long and could be greatly condensed. At first glance and speaking from how I like to lay things out in a paper:

- You do not really need to show the prompt interface. You can show it once and thereafter use a bulleted list format, or simply show the input image if it responded correctly. - Your figures should be about half of their size, they don't need to fit the width of the body. - For comparative results against other models you can use a table with colored cells, with the test name on the row and model names on the columns. - For the dog, show a side-by-side figure with the raw image on the left and the box on the right, and include the coordinates it gave you in the body. - In your conclusion show the full matrix table of comparative results and summarize the relative strengths of the model against the others.

In terms of the writing and methods your conclusion says little and your tests do not go into significant depth. For example with the tire image you could show that it succeeds when cropped but as the photo gets wider it begins to fail to correctly identify the text in the image's center. For example see the methodology and presentation this article used: https://dynomight.net/ducks/

Also, the OCR test is too simple, even a 20-year-old OCR algorithm would probably recognize that. Experimenting with progressive degradation of the image could show its strengths, and…

It is shockingly difficult to use LLMs in chess study. I don't need it to be a better (or worse) Stockfish; an LLM should be great at taking a FEN or multiple lines from Stockfish via MCP or tool call and explain why positions are evaluated the way they are, typical plans in the position (drawing from pretraining knowledge of a vast archive of games), and how to explain to a human to study these positions.

I suspect that the large amount of chess pretraining data is not well synchronized with positions, because in books and articles the text is typically accompanied by pictures of the positions, NOT FENs / PGNs. So the training on the text is decoupled from the representation of the position.

Regarding your tool call thing with stockfish/maia, I made a tool like this for myself called Blunder Sniper which iteratively feeds positions - that I'm likely to get given my openings - and recursively calls the lichess DB and finds the first time in the top 80% played moves in each chain where the opponents in the rating range blunder as the most common move in the chain.

It was a fun way to use an alternative to engine-based preparation that many strong players use, which is something like Nibbler + lc0 and using contempt high values to find higher variance lines rather than game theory optimal ones.

Some day I'll expand on the gpt-chess articles [0] that I found super interesting, fine-tune models... well, I keep telling myself that, anyway...

[0]:…

You can indeed take this route [0] and even DIY something with a box fan. But I think the jury is a bit out on this. Quote from that link below. You also need to move that much more air to get the multiple passes required and that can get pretty noisy.

> They claim that around 90% of particles sized 0.3 microns are larger [for a MERV] are eliminated in a single pass. That’s good, but not totally reassuring. The question is, does it remove 99% in two passes? If 90% of the particles in the ambient air were large and the filter only catches large particles, then additional passes would never get rid of the most dangerous small particles. This is why I trust HEPA filters a bit more: since they remove almost all particles in one pass, I’m confident they should remove almost all particles eventually.

Also, there is the idea that HEPA filters act like a sieve. They don't. This article from the same guy is a great read [1]. So basically, multiple passes for the win.

> Air filters easily catch both large and small particles. It’s the intermediate regime where things are hard. The size where the filter performs worst is called the Most Penetrating Particle Size (MPPS). Typically this is around 0.15 microns.

[0] https://dynomight.net/2020/12/15/some-real-data-on-a-DIY-box-fan-air-purifier/

[1] https://dynomight.net/ikea-purifier/

> My real worry about seed oil theory is that it’s a distraction. If you want to be healthier, we know ways you can change your diet that will help: Increase your overall diet “quality”. Eat lots of fruits and vegetables. Avoid processed food. Especially avoid processed meats. Eat food with low caloric density. Avoid added sugar. Avoid alcohol. Avoid processed food.
>
> I know this is hard. You could even argue it’s unrealistic. That wouldn’t make it wrong.
>
> Look, I wish strong seed oil theory were true. That would be great. All we’d have to do is reformulate our Cheetos with different oil, and then we could go on merrily eating Cheetos. Western diet without Western disease! Sadly, I think this is very unlikely.

Dynomight has written an excellent article on seed oils. We have examples in our families of people with very bad diets, favouring tons of butter because ‘at least it’s not processed’.

The research doesn't back this up. As the quote above suggests, we need to stop thinking of ‘magic bullets’ that’ll get us out of eating better. Added sugar and trans-fats might be the only true bullets.

https://dynomight.net/seed-oil/

My phrasing was perhaps too strong. I learned this from here https://dynomight.net/air/
This is a great read.
Here is a previous HN discussion on this:
https://news.ycombinator.com/item?id=26964722
There are some disagreements but I find the argument that air quality is greatly underappreciated as a health risk convincing. And that candles cause bad air quality.

The article talks a lot about healthspan which I think is a great measure and idea. Your referenced article talks about at-risk people and that they should avoid candles. It also states that candles should be burned in well ventilated spaces, which I believe most people don't do as they don't consider it.

A healthy persons healthspan is negatively affected by candles and such. They just have more health to spare.

The article talks about cooking as well and they are likely talking about cooking with gas which is very unhealthy. It should be illegal to sell indoor gas stoves for cooking in my opinion.

> Robert Lustig videos on Youtube are a great way to start learning about this.

Or off YT:

https://videos.lukesmith.xyz/w/feioCyaQEZ1ogHK1oNJM4K

> Thus, the people taking fiber out of and putting sugar into everything are the biggest current mass killers. Stringing them up may be an over-reaction. Maybe convicting them would be, too? Anyway stopping them seems like a good idea. Think about them next time the news is full of somebody killing a dozen people. Is a dozen bad, but a million A-OK?

IMO simply not compelling millions of people to pay them for it would be a good start. Defund all the agencies that promote poison, whether by subsidizing corn syrup, villianizing natural saturated fats to sell trans fats and sugar, or conspiring with drug companies and corrupt researchers to sell addictive substances (as discussed this week: https://dynomight.net/alcohol-trial/).