Hacker Newsnew | past | comments | ask | show | jobs | submit | ChemSpider's commentslogin

"insulin resistance trends" - whaaaaat? how would they do THAT? That would be huge for so many people, including diabetics, obese people or women with PCOS (PCOS alone affects more than 10% of women of reproductive age)

Is that measurement accurate? Like, a person starts taking Metformin -> and a few days later the watch detects a drop in insulin resistance?


No, it's not a glucose sensor. Extrapolating from the Google research published in Nature a few months ago on how they estimate insulin sensitivity using wearables (and not just BMI, age, etc): https://www.nature.com/articles/s41586-026-10179-2

Insulin sensitivity correlates with obesity, activity, sleep, etc. so it's not impossible that a wearable can give some degree of prediction. I guess it's mostly good motivation to support habits that increase sensitivity (physical activity, low stress).


FDA compliance has been tough for optical measurement of blood sugar and blood pressure. I'd imagine that this is why they are doing a monthly averages and keeping their claims subdued.


Yeah I find this one interesting. It's been rumoured for pretty much every Apple Watch for a decade, so they're clearly throwing tonnes of resources at it, and every year nothing materialises. This is one sensor feature that could really impact a lot of people.


Definitely not. Get a cheap CGM subscription if this is something you want to monitor and you need any sort of real data. They’re easy to apply, update quickly (within 15 minutes of a Hcg spike you will see it in the data) and you can derive A1C from it. Actual insulin resistance itself still requires a fasted blood draw at minimum and is not really recommended (too many methods with different ways of measuring each of which is clinically significant but needs interpretation — or you can do a 3-4 hour research lab test but the value is pretty limited).

Any time you see “trends” in a press release you should be very suspicious. It’s a sign that they can’t pass regulatory scrutiny and are trying to apply the research anyway. Your scale cannot measure fat percentage, your Samsung watch cannot measure blood pressure and your Garmin cannot measure VO2Max. These are all things they’d like to be able to do (and “trends” can have some value if they’re used correctly) but they are not the thing itself.


I'm quite curious. Friends I've talked to have been very pessimistic about accurately measuring blood sugar via a watch.

If they are actually able to get an optical measurement which is good enough for MoM or WoW trends, I'd agree it's pretty big even if they can't provide anything precise enough for real time / daily use. Just being able to roughly validate diet changes for women with PCOS could be very impactful and that's the smallest of the 3 populations you called out.


It does monthly trends: https://m-cdn.phonearena.com/images/articles/452350-image/in...

That image also implies that not only does it know trends, but it also has an absolute good/bad value.


Yeah, I was researching this just last week and there is tons of funding going into wearable blood sugar monitors but nothing wearable sized coming out (several lab sized ones though).

I’d love to read more about what this is, and what technical signals / data they’re using.


>(because you shouldn’t move a small town if you are non white)

Where does this rule come from? It does not match my experience.


"Early 2025: First enterprise contracts secured" "Today: Powering Fortune 500 engineering teams" - I guess that is all that is publicly available for now.


Built what you want to use yourself. AI for engineering and physics sounds like the perfect product a company like ASML (Mistral investor) could use.


Yes and they are open about it. That is why they are building their own data centers in Paris and other places.


That’s true, but as their model quality is behind Kimi K2.6 or Deepseek v4, I believe it makes more sense to consume those Chinese models from a European hosting company like Scaleway, than to consume Mistral’s models hosted on US cloud providers, except if you see that Mistral subscription as a kind of investment into Mistral for them to buy infra in the future. That is, if your goal is to “consume” AI on European soil.


Behind Deepseek R1 by far, they are at least 1 year late in model's performance.


What does a Physics prof at Berkely earn these days? In Germany it would be around 10K€/month (lets ignore all tax etc stuff, just ballpark figure)


€10K seems high to me. In France it would be lower than that, I think, save for maybe a handful of outliers. Google search suggests top earners in physics in academia in Germany is probably between €80K and €90K, or maxing out at €7.5K a month.


About 200k USD, I think? But isn’t 10k€/month relatively high for europe, like on the upper end of academic salaries?


The challenge with CO2 monitoring is the sensor, not the electronics. Sensor accurracy and service life are key information.

It is easy to create a low power chemical CO2 sensors with a service life of a few weeks/months. Obviously not pratical for real world applications. So critical data is missing in this press release.


Replacing sensors every few weeks is totally practical for some real-world applications such as life support systems.


You think of CEO salaries. PhD students and PostDocs earn more (or at least equal) in most of Europe compared to even the US top universities.

Harvard Postdoc salary: $67,600

Germany Postdoc salary: ~$80,000 plus benefits like health insurance


That is hysterical. You know that salaries for most universities are public in Germany? You can just look up your ridiculous lies.

https://vertretungen.hu-berlin.de/de/personalrat/tarif

The highest possible salary anyone in the university gets is 96k Euros. That is 64k Euros a year after taxes.


Did you actually read the page you linked??? It supports exactly what I wrote. Just keep in mind that the latest salary data there is for 2022.

The postdoc salaries I quoted are for 2024, of course.

Source: https://www.research-in-bavaria.de/what-salary-does-a-postdo...

"Most doctoral positions and some postdoc positions will be categorized as TV-L 13, which can range from about €4630 to €6580 (gross monthly salary). The exact salary is determined by your years of experience."

-> €6000*12 -> 72K Euro, ~US$ 80K

And that is without including the various social benefits.

> The highest possible salary anyone in the university gets is 96k Euros.

That is wrong, too. But not the topic of my post.


>Just keep in mind that the latest salary data there is for 2022.

Haha. This is NOT data, this is the current contract. It is exactly what you get paid. It is the exact salary you get paid there.

>Source: https://www.research-in-bavaria.de/what-salary-does-a-postdo...

Why don't you look at the document I linked which tells you exactly what people working in that university are earning. That is not a guess, not an estimate and it is not out of date it is literally exactly what they are getting paid next month.

Instead of talking about hypotheticals and ranges, why do you refuse to look at the document which talks about the exact salaries? This is very bizarre behavior.

>That is wrong, too. But not the topic of my post.

I literally linked you the literal document where it is literally written what the literal highest paid person will literally get paid in the literal next month. What evidence against this could you possibly have? Unless you have the bank statements of that person your evidence cannot possibly be better.


The link your parent posted is current. The Entgelttabelle there was last updated in 2022.

Also the PDF's (Tarifvertrag, Entgelttabelle) linked there are the authoritative source for this information - they are the legal document defining it.


We have 2025, not 2022. So it is certainly not current. Inflation is a thing! ;-)

My data (see above) is from the same source ("Tarifvertrag"), but from 2024.

I am really surprised that PhD/PostDocs earning more (or at least same) in EU than the US is controversial. I tought everyone knows that.


>We have 2025, not 2022. So it is certainly not current. Inflation is a thing! ;-)

Do you not understand that the document I linked is the up to date contract? If you think what I linked is getting "inflation adjusted" you are totally clueless about how you get paid under a Tarifvertrag in Germany.

The document contains the exact amount of the employee is going to get paid next month.


You are good with insulting, but weak on math.

Even with your 2022 data/source, I get a salary that is >=Harvard.

And if you look at the latest 2025 salaries the difference is even more striking and in favor of Germany:

https://www.jobs-beim-staat.de/tarif/tvoed-bund_e13

But according to you, how much does a German Postdoc earn per year, in US$?


First year 50,260.56 Euros at TV13 at the Humboldt University Berlin compared to Harvard's MINIMUM salary of $67,600. See https://postdoc.fas.harvard.edu/salarystipend-level

Given that taxes are much, much higher in Germany the difference is quite drastic.

Again I am NOT using a source for 2022, it is the exact contract in force for May 2025.


"it is the exact contract in force for May 2025." - No, it is not. That is your mistaken thinking.

In 2025, TV-L 13 starts with 4629€ a month and goes up to 6580€ (= 78K€+), depending on experience.

Indded, the raw tax difference is substantial. But the social benefits, health insurance, vacation days,... need to be considered for a complete comparison. But that was never the topic of discussion.


I demonstrated to you that the starting salary differs by 18k, at 1:1 exchange rate. I did this with up to date information that is valid RIGHT NOW.

>In 2025, TV-L 13 starts with 4629€ a month and goes up to 6580€ (= 78K€+), depending on experience

Wrong. You do not understand how TVs work. Look at the linked table it as the exact salary the employees get paid at the end of this month.

>depending on experience.

Wrong. Experience is totally irrelevant. It is years in that position.


You are completely delusional. As you have been told, even when evaluating favorably the EU job, the difference will be so substantial that it's hardly comparable.

You may think there are many advantages of the social system, but the thing is with the US you get to choose exactly how the surplus will be spent. If you want coverage similar to EU you can, it will cost you lots of money but you will still come out on top.

And that is before even considering that purchasing power/quality of life is better in the US at a given salary level.


The controversy is your egregious dismissal of a legal source that you have now edited out of your post.


So you are saying the average German Postdoc salary is NOT >= Harvard salary?

What numbers do you have in mind?


Unfounded provocation. Get lost, troll.


It is about grant money, not salaries.


500 US$ = 442 euro. Does that include the higher doses? Than it is actually cheaper than in Europe.

15mg Zepbound is 489 euro in Germany for a month (4 shots).


$499 per month "[f]or the Zepbound 5 mg, 7.5 mg, and 10 mg vial" [1]. So, I don't know how much they charge for 15 mg.

[1] https://zepbound.lilly.com/coverage-savings


"World's best OCR model" - that is quite a statement. Are there any well-known benchmarks for OCR software?


We published this benchmark the other week. We'll can update and run with Mistral today!

https://github.com/getomni-ai/benchmark


Update: Just ran our benchmark on the Mistral model and results are.. surprisingly bad?

Mistral OCR:

- 72.2% accuracy

- $1/1000 pages

- 5.42s / page

Which is pretty far cry from the 95% accuracy they were advertising from their private benchmark. The biggest thing I noticed is how it skips anything it classifies as an image/figure. So charts, infographics, some tables, etc. all get lifted out and returned as [image](image_002). Compared to the other VLMs that are able to interpret those images into a text representation.

https://github.com/getomni-ai/benchmark

https://huggingface.co/datasets/getomni-ai/ocr-benchmark

https://getomni.ai/ocr-benchmark


Do you benchmark the right thing though? It seems to focus a lot on image / charts etc...

The 95% from their benchmark: "we evaluate them on our internal “text-only” test-set containing various publication papers, and PDFs from the web; below:"

Text only.


Our goal is to benchmark on real world data. Which is often more complex than plain text. If we have to make the benchmark data easier for the model to perform better, it's not an honest assessment of the reality.


Excellent. I am looking forward to it.


Came here to see if you all had run a benchmark on it yet :)


It’s interesting that none of the existing models can decode a Scrabble board screen shot and give an accurate grid of characters.

I realize it’s not a common business case, came across it testing how well LLMs can solve simple games. On a side note, if you bypass OCR and give models a text layout of a board standard LLMs cannot solve Scrabble boards but the thinking models usually can.



Interesting. But no mistral on it yet?


Its Mistral, they are the only homegrown AI Europe has, so people pretend they are meaningful.

I'll give it a try, but I'm not holding my breath. I'm a huge AI Enthusiast and I've yet to be impressed with anything they've put out.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: