zurfer

Reliability is the important dimension to focus on, but what is your baseline? I've worked in data and me and my colleagues regularly had to confront bugs (on many levels) that were communicated to end users

Germany is the 5th biggest economy in the world, then there is Austria and Switzerland. The claim that nobody would pay for German to SQL seems a bit pessimistic ;)

I'd also love to understand better why you think that there is no "acceptable level of value-add in most reasonable enterprises".

Cofounder of one of those analytics agents here (https://getdot.ai).

The promise of the technology is not that it can deal with any arbitrarily complex Enterprise setup, but rather that you expose it with enough guidance on a controlled and sufficiently good data model.

Depending on your use case this can be super valuable as it enables a lot more people to use data and get relevant recommendations.

But yeah it's work to make nice roads and put up signs everywhere.

Regulation might be one part of the equation, but I would like to understand a bit deeper on how much of cost driver it is.

Healthcare is heavily regulated in some countries and less in others. It should be possible to get some comparisons.

I somehow feel that making many humans healthy is fundamentally a really hard problem that gets harder every year because the population ages and expectations rise. It feels to easy of a talking point to put it all on regulation.

Commercializing something gives you more resources to do the thing you really want to do.

Anthropic came to the same conclusion. SSI either doesn't have something commercializable or came to a different conclusion.

Commenting on such a broad and general theme as the article does feel useless. Bureaucracy yadada, people standing on an empty street light... sausage is ok.

Doesn't speak the language, claims the people are not free. How shallow.

There are more interesting models to build here. More interesting art to create. But hey it provoked me. Like most modern media, it made me a bit sad.

Horrible webpage. No clear CTA. No pricing page. Not even "talk to sales". How am I supposed to get started? /s

Well you wouldn't bet all your assets because it would be an illiquid market that could only resolve in your favor in earliest 80 years.

If you're really serious about it put the money into a prediction market. Poly market has multiple AGI bets.

Yes it matters because Google ad spend is just one way to market and it's harder to attribute sales if there is a lot of fraud making it more inefficient.

You're both kind of right. The order is less important for reasoning models, but if you carefully read thinking traces you'll find that the final answer is sometimes not the same as the last intermediary result. On slightly more challenging problems LLMs flip flop quite a bit and ordering the output cleverly can uplift the result. That might stop being true for newer or future models but I iterated quite a bit in this for sonnet 4.

[flagged]

Inference etc should go in this bucket: "Operating losses reached US$7.8 billion"

That also includes their office and their lawyers etc , so hard to estimate without more info.

The $13.5B net loss doesn't mean they are in trouble, it's a lot of accounting losses. Actual cash burn in H1 2025 was $2.5B. With ~$17.5B on hand (based on last funding), that’s about 3.5 years of runway at current pace.

There is another under discussed alternative UHV power transmission, e.g. south to north: Morocco has great conditions for solar. Or East to West, the sun rises and sets at different times.

We still need more storage and generation, but a better grid would help a lot.

Also getting a perfect score on AIME (math) is pretty cool.

Tongue in cheek: if we progress linearly from here software engineering as defined by SWE bench is solved in 23 months.

Same price and a 4.5 bp jump from 72.7 to 77.2 SWEBench

Pretty solid progress for roughly 4 months.

Actually bad luck, it seems this doesn't support Microsoft Fabric with the datawarehouse engine... it fails because Fabric doesn't support the DECLARE CURSOR operation, which this driver relies on.

This is really timely. I just needed to build a connector to Azure Fabric and it requires ODBC 18 which in turn requires openssl to allow deprecated and old versions of TLS. Now I can revert all of that and make it clean :)

"The report concluded that Israel has committed genocide against Palestinians in Gaza since 7 October 2023, covering the period from that date until 31 July 2025.

It said that Israel has committed four acts of genocide:

    Killing members of the group: Palestinians were killed in large numbers through direct attacks on civilians, protected persons, and vital civilian infrastructure, as well as by the deliberate creation of conditions that led to death.
     
    Causing serious bodily or mental harm: Palestinians suffered torture, rape, sexual assault, forced displacement, and severe mistreatment in detention, alongside widespread attacks on civilians and the environment.
     
    Inflicting conditions of life calculated to destroy the group: Israel deliberately imposed inhumane living conditions in Gaza, including destruction of essential infrastructure, denial of medical care, forced displacement, blocking of food, water, fuel, and electricity, reproductive violence, and starvation as a method of warfare. Children were found to be particularly targeted.
     
    Preventing births within the group: The attack on Gaza’s largest fertility clinic destroyed thousands of embryos, sperm samples, and eggs. Experts told the commission this would prevent thousands of Palestinian children from ever being born."

Genocide can't be measured by intend, because we can't look into someone's head. It's measured by the actions that are taken. And while I do agree that Israels actions are a mixed bag, I feel too many lines are crossed to assume only good intend.

It's a basic need for people to feel safe. I wish that for everybody and most of all for the children of this world.

Legal judgements often make it to the front page of HN as they are as independent as we manage as humans. I don't feel having this post slanders Israel. It would be more interesting to understand what part of the UN investigation you disagree with.

It was such a great study imo and rightfully deserves a lot of attention since it's a great alternative to twitter vibes.

But to stop here and say AI coding doesn't work will just not hold up well. We have a sample size of 16 developers for 250 tasks.

Here is my data point: adding AI to your engineering team really is like adding a new member. At first it slows everybody down: explain this, explain that, stupid mistake made that the experienced dev would have avoided. But over the months you learn how to get the most out of them: they speak fluently all programming languages, don't mind writing 20 tests or coming up with good industry ideas to your problems.

Over time I have come to appreciate that the market is rather smart and even hype has its place. You often have to cross a line to know you crossed it, especially with something as fuzzy and quickly changing as technological progress hitting the messy real world. But then again, we need the heretics to stay smart and not follow an emperor without clothes.

Thanks for sticking the head out ;)

I'm really not sure where 1 to 3 come from? 1 and 2 are about spending time with people. 3 can't also be true, just look at all the innovations that OSS has created (at least 90 perc. remote collaboration).

Number 4 is a real problem and a symptom of the biggest problem imo, which is that it's hard to build trust online/remotely.

Maybe. On the one side you have something like dbt or Moosestack. On the other hand analytics and data pipelining is still a lot of no code tooling and I doubt it will go away. However I would love to learn more about how other people use coding agents to do DE tasks.

Woah that's really kind of hidden. But I think you can specify max output tokens. Need to test that!

He probably heard the memes, didn't really want to work for Meta and said a "ridiculous" high number and Meta was like: we can do that.

The simple reason for this is that Claude Code uses way more context and repetitions than what you would use in a typical chat.

Little nitpic, Gdansk was a free city, neither Polish nor German when nazis annexed it in 1939. Most of the destruction happened in 1945 because of fighting with the red army.

Unfortunately, I only got data until 2022, but here is a similar overview with a few more charts and sentiment analysis: https://eu.getdot.ai/share/f3f0853d-fa91-4301-8fb2-52821b65e...

Will try to update it with some more recent data later.

Everybody talks about what this does to OpenAI, but I do wonder how this shakes out for Meta.

If the person next to you gets paid 20x more than you, you might be a bit unhappy when they are not 20x more helpful.

"Rick likes books from Tufte. Tufte is known for his work on data visualization. Is Rick interested in data visualizations?" (all frontier reasoning models get that right).

-> This qualifies for me as a super simple reasoning task (one reasoning step). From that you can construct arbitrarily more complex context + task definitions (prompts).

Is that "just" statistical pattern matching? I think so. Not sure what humans do, but probably you can implement the same capability in different ways.