bethekind

I love the phrase "inference hunger"

Jane Street had a cool video about how you can address lack of training data in a programming language using llm patching. Video is called "Arjun Guha: How Language Models Model Programming Languages & How Programmers Model Language Models"

The big take away is that you can "patch" llms and steer them to correct answers in less trained programming languages, allowing for superior performance. Might work here. Not a clue how to implement, but stuff to llm-to-doc and the like makes me hopeful

Model interpretability is going to be the final frontier of software. You used to need to debug the code. Now you'll need to debug the AI.

This is draconian.

> Our investigation specifically confirmed that the use of your credentials within the third-party tool “open claw” for testing purposes constitutes a violation of the Google Terms of Service [1]. This is due to the use of Antigravity servers to power a non-Antigravity product. I must be transparent and inform you that, in accordance with Google’s policy, this situation falls under a zero tolerance policy, and we are unable to reverse the suspension. I am truly sorry to share this difficult news with you.

Did we read the same graph? DeepSeek Gundam 200 dpi appeared to get similar perf as dots-ocr, but with less tokens needed. The x axis is inverted, descending with distance from the origin.

> the behavior of a complex system can simplify as it passes from one state to another. “Sometimes a high-dimensional system can tip,” said Lenton, “and when it gets near tipping, it starts to behave like a much lower-dimensional system.” The lesson, he added, echoes the one learned at Peter Lake: to “simplify without oversimplifying.”

Sounds a lot like what people WANT neural networks to do. Collapse a high dimensional situation into a very low dimensional network. Ideally a binary answer, yes or no. I wonder if this bifurcation chaos math has implications in ml work

If Eli Lily is the only producer, how is the gray market being supplied? This makes no sense...

That being said, I'm waiting for oral GLP1 agonists. Injections are a hassle and gray market ones even more so

> ...as much as it is for anyone....

The phrasing in this sentence implies that the Julia language could be named for the code breaker, as much as it could be named for anyone else. In other words, it wasn't named for the code breaker, but it might as well have been.

The follow up comment gives hard quantitative fact that the language wasn't named for anyone or anything. I can see how both comments are correct, the first implicitly, the second explicitly

This is basically a mini grow light right?

I think a good rule of thumb is to stop if it hurts. I presume you're not using a laser. Your eyes are pretty good at telling you when something is too much

I think one of the issues is how over priced most of the panels are. You're paying an arm and a leg for lots of little red lights. Nothing more.

I bought an infrared parabolic heater to test this out once, as a friend made a comment insinuating that more time in front of fire meant a better sleep. I ran it for a few days, but the results were mixed and inconclusive.

I suspect because the longer IR wavelengths just heat the body up, it's hard to get enough of the more "useful" wavelengths without overheating

Sun is good, but has limitations

It has UV, which stresses out the cells You can't PWM the sun, it's hard to get full naked coverage in front of it. Doesn't perfectly target NIR or NIR-II biological windows.

But for a base/starter, it can't be beat.

Simply put, light is better than no light.

Most of the stuff online is overpriced and beyond the mitochondrial enzyme absorption spectrum, I can't find any reasonable mechanism for action.

I think your hesitance is valid

I think the point of wanting a control is still valid. Science always benefits from having a control

It's quite analogous, especially for ideal intensity

Bright enough to warm the body, but not thermally stress the cells.

And yeah, closing your eyelids would block UV light from getting in and cross linking proteins and causing cataracts, while still allowing for NIR light to get in and speed up the mitochondria.

Analogous for sure

Build your own. But direct from Alibaba or mouser or digikey.

The main science behind it is to hit the absorption peaks of a mitochondrial enzyme. You can make em bright, but be limited on how long you can use it, or dim and be forced to use it constantly for the same effect.

Pulse width modulation is very important in this for penetration. High intensity allows for deeper penetration, but heats up cells causing stress. The ideal scenario is a really strong light, run at a really low duty cycle, allowing for both intensity AND low average wattage, this lessening thermal issues

There might be other mechanisms of actions, as there are so many papers that say "light good", but until they can give a mechanism of action, I'm sticking with what I know

I have so many questions about both apparatuses. 90W input or irradiated? Pulse width modulated/dimmed? Lasers? LEDs? 850nm? 830nm? 810nm?

For the panel, 1800W is a LOT of power to put through 2 feet. Is it actually 1800W? What wavelengths? PWM?

I've been using a NIR belt flipped inside out on my pillow the last few weeks. It's only 6W of 850nm, but I've been feeling less dumb recently. Not sure if it's correlated, but until I settle it for sure, I'm going to keep on using it.

Do you use llama 3 for your work?

Are drones like the reaper easier and cheaper to maintain than normal planes like the f-35?

I could easily see USA releasing hybrid ships, with a few jets, but lots of drones

Wattage might not be high enough to notice a difference, but might be fun

A YouTuber calls thought emporium tried out the whiskey method

And yes, ultrasonic cavitation will break apart the grounds very thoroughly. Think a jewelry or denture cleaner.

This had me cackle :D

Using low frequency sound will agitate the solute and solvent, speeding up the natural cold brew process.

Using ultrasonic will do it even faster, but since ultrasonic underwater induces cavitation bubbles, it's much more violent.

Ultrasonic has been used for dermal infusion quite successfully, but it is....painful, as bubbles are exploding against your skin.

I would presume the same to be for the sonic coffee. Agitation speeds up the process, until cavitation occurs, where it becomes more violent

Robot taxis not coming soon enough?

Profitable, yes but pennies per day. 5950X should make ~$5/month after all is said and done iirc.

If you live in cold climates where space heaters are used, it makes sense, as you would've been heating the house anyways.

I've always wondered how the epycs with huge amounts of L3 would perform on monero.

Is there an optimal core to l3 ratio? Or is it always more is better

Corporate entities, which followed the calendar instead of listening to the trees probably didn't do well this reason

Multiple people on Reddit said the same thing, that this year was one of the best for small time shops.

I think this article is part FOMO, part "oh no feel bad for us" by big companies.

Could you expand on what you mean by "practitioner?"

Also a followup question. With timeGPT and chronos advertised as "foundational time series models", do you think they have any value?

I played around with timeGPT beta against predicting the sp500 index performance for the next day (not multi variate time series as I couldn't figure out how to get it setup) and trying to use the confidence intervals it generated to buy options was useless at best

I can see chronos working a bit better, as it tries to convert trends, and pieces of time series into tokens, like gpt does for phrases.

Ie. Stock goes down terribly, then dead cat bounces. This is common.

Stock goes up, hits resistance due to existing sell orders, comes down

Stock is on stable upward trend, continues upward trend

If I can verbalize these usual actions, it's likely chronos can also pickup on them.

Once again quality of data trumps all for LLM's, so performance might vary. If you read the paper, they point out a few situations where the LLM is unable to learn a trend, ie. When the prompting time series isn't long enough.