HNHacker News
TopNewBestAskShowJobs

carlmr

7,934 karma · joined September 1, 2017

submissionscomments
carlmr··on Java is fast, code might not be
The problem is rather that Java doesn't have generics and structs, so you're kind of forced to box things or can't use collections.
carlmr··on Tesla: Failure of the FSD's degradation detection system [pdf]
Nice. My test was always a blond bald guy. It always adds hair. If you ask for bald you get a dark haired bald guy, if you add blond, you can't get bald because I guess saying the hair color implies hair (on the head), while you may just want blonde eyebrows and/or blond stubble.
carlmr··on Tesla: Failure of the FSD's degradation detection system [pdf]
Not unlikely that you're talking to a lot of AI-based AI boosters. It's easier to create astroturfed comments with chatbots than fixing the inherent problems.
carlmr··on Tesla: Failure of the FSD's degradation detection system [pdf]
I'd add to this, any moderately involved logical or numerical problem causes hallucinations for me on all frontier models.

If you ask them in isolation they may write a script to solve it "properly", but I guess this is because they added enough of these to the training set. But this workaround doesn't scale.

As soon as I give the LLM a proper problem and a small part of it requires numeric reasoning, it almost always hallucinates something and doesn't solve it with a script.

If the logic/math is part of a larger problem the miss rate is near 100%.

LLMs have massive amounts of knowledge, encoded in verbal intelligence, but their logic intelligence is well below even average human intelligence.

If you look at how they work (tokenization and embeddings) it's clear that transformers will not solve the issue. The escape hatches only work very unreliably.

carlmr··on Tesla: Failure of the FSD's degradation detection system [pdf]
>I definitely agree that in principle a computer can drive with cameras alone.

Obvious things first, cameras have way worse contrast and low light sensitivity than human eyes.

Humans have much more evolved logical thinking capacity, even the stupid ones can figure stuff out that modern AI struggles with.

Humans have other sensors, too that they use to plausibility check the picture they see. I.e. one of the best sensor fusion systems on the planet.

When in doubt humans can figure out whether it's a lens occlusion or a some other artifact in their vision by virtue of moving their head around.

There's probably other things I'm not thinking of. In any case to make full self driving work we should first start by using all available tech to make it safe. When you have safe tech you can slowly start removing individual sensors while verifying that safety remains high. As the experience and system evolves there will be optimization potential.

And until we have that low light thing and high contrast figured out, camera alone doesn't cut it.

carlmr··on A sufficiently detailed spec is code
Exactly, you would probably also pay for a good one. I tried maybe 10 and gave up. I now carry a small booklet where you can rip off the pages, for notes and todos. Still way better than the phone apps
carlmr··on A sufficiently detailed spec is code
Which is not getting better.

I'd pay you 10€ for a TODO app that improved my life meaningfully. It would obviously need to have great UX and be stable. Those are table stakes.

I don't have the time to look at all these apps though. If somebody tells me they made a great TODO app, I'm already mentally filtering them out. There's just too much noise here.

Does your TODO app solve any meaningful problem beyond the bare minimum? Does it solve your procrastination? Does it remind you at the right time?

If it doesn't answer this in the first 2 seconds of your pitch you're out.

carlmr··on Beyond has dropped “meat” from its name and expanded its high-protein drink line
I think the problem is lot's of people here don't have much kitchen experience and underestimate the effect.

But anyway, I think a pre-seasoned vegan ready made burger patty should only be compared to a pre-seasoned meat burger patty. It's an Apples and Oranges comparison with little meaning.

If you compare the high sodium of a vegan ground beef replacement with ground beef, that's fair game. The one from Beyond here is actually a good example of too high sodium. I won't judge. I only care about the comparison, not the company.

carlmr··on Beyond has dropped “meat” from its name and expanded its high-protein drink line
Thanks, so the squasher is all I need I guess
carlmr··on Beyond has dropped “meat” from its name and expanded its high-protein drink line
> I got a kit for making my own burgers from dehydrated soy mince and/or mashed kidney beans.

Do you have a link or name for this? I also prefer black bean or lentil burgers, but I've been making them by hand really.

carlmr··on Beyond has dropped “meat” from its name and expanded its high-protein drink line
>it has way more sodium than ground beef you'd buy at a grocerty store

We're not comparing fairly here. A finished hamburger patty is not pure ground beef. Did you ever make a hamburger patty yourself? You add salt and spices at a minimum.

A more fair comparison would be looking at store-bought hamburger patties. That's the same category of food.

I just compared Beyond (0.75g salt per 100g) and block house American Burger (0.88g per 100g). The patties are somewhat similar in weight, too (113g and 125g). So both in absolute, and weight relative amounts the Beyond burger has less sodium.

carlmr··on Grief and the AI split
This might be happening, too, however B2C advertising and heavy astroturfing is a sure sign that they don't even think they're close to this goal.

The average consumer pays the least for subscriptions and asks most uninteresting questions to the AI in terms of gaining insight. The only goal here can be upholding the narrative that everything will be AI soon™.

carlmr··on The AI coding divide: craft lovers vs. result chasers
>But instead, all you see are FOMO propaganda to get devs to adopt the tool with no asking if it actually helps the devs do their job.

If (big if) LLMs/AI take over all of knowledge work the first thing you'll notice is that the first company getting to the point of automating all knowledge work will close off their models to the public, not advertise it, and take over every business on the planet.

You wouldn't waste a dime on advertising, influencers, or convincing people to use your product.

Taking over every business in the world seems more lucrative than selling $20 subscriptions to people.

carlmr··on Amazon Employees Say AI Is Just Increasing Workload
>We have strong KPIs on how much we use copilot so I must say woof woof to it every day or so to make sure it shows me as an active user (luckily it only shows the latest date someone has used it in each application).

Ask it to generate a cron job for this.

carlmr··on Yann LeCun raises $1B to build AI that understands the physical world
Especially they will require even more compute to get anything close to usable output. Human brains are super efficient at learning and producing output. We will need exponentially more compute for real time learning from video + audio + haptic data.
carlmr··on How Much Money Jeff Bezos Made Since You Started Reading This Page
I think idealists often get into politics as well, but they're not cold, calculating and power hungry enough to get into the important positions.
carlmr··on Mullvad VPN: Banned TV Ad in the Streets of London [video]
>it becomes up to a body of bureaucrats to determine what is true.

I think we have a ministry for that.

carlmr··on MacBook Pro with M5 Pro and M5 Max
I've upgraded to Tahoe at 26.2, zero complaints from my side. Haven't had any runaway memory leaks or similar that were reported.
carlmr··on Motorola announces a partnership with GrapheneOS
There are MacBook companies, but Lenovo is basically the only other alternative.

I've had Dell in the past, but haven't seen one in years.

carlmr··on What does " 2>&1 " mean?
It's not necessarily a leaky abstraction. But a lack of _knowledge in the world_.

The abstraction may be great, the problem is the lack of intuitive understanding you can get from super terse, symbol heavy syntax.

carlmr··on Google API keys weren't secrets, but then Gemini changed the rules
It's never that explicit, it's more the things that nobody takes care of, because it's nobody's job. The bigger the company, the more jobs fall through the cracks, that should be taken care of, but lack an explicit role in the hierarchy.

There's usually a small handful of people that care more than they should, keeping the company afloat, but it's despite the company's policies, not because of them.

carlmr··on How will OpenAI compete?
But everyone had at least one friend, Tom.
carlmr··on Google API keys weren't secrets, but then Gemini changed the rules
If you've never worked in a large corporate environment you don't know how stupid things become. In a perfect bureaucracy nobody thinks.
carlmr··on Halt and Catch Fire: TV’s best drama you’ve probably never heard of (2021)
It’s funny, I’d say the details are right, but the overall picture is still wrong.

It tries to cram too many things into this one show. Like a medley of computing history.

carlmr··on Learning from context is harder than we thought
To stretch the human analogy, it's short term memory that's completely disconnected from long term memory.

The models currently have anteretrograde amnesia.

carlmr··on Coding agents have replaced every framework I used
I was thinking the exact same thing. Moltbook isn't that sophisticated. We're moving goal posts a lot here.

However, I do think 1 week is ambitious, even for a bad clone.

carlmr··on LLMs as the new high level language
>IIUC, same model with same seed and other parameters is not guaranteed to produce the same output.

Models are so large that random bit flips make such guarantees impossible with current computing technology:

https://aclanthology.org/2025.emnlp-main.528.pdf

carlmr··on How to effectively write quality code with AI
>They say that using LLMs as advisors is already outdated, we should be doing fully agentic coding and just nudge the LLM etc. since we're losing out on 'productivity'.

As long as "they" are people that either profit from FOMO or bad developers that still don't produce better software than before, I'm ok ignoring the noise.

carlmr··on How to effectively write quality code with AI
>but the code to me is a forcing mechanism into ironing out the details, and I don't get that when I'm writing a specification.

This is so on point. The spec as code people try again and again. But reality always punches holes in their spec.

A spec that wasn't exercised in code, is like a drawing of a car, no matter how detailed that drawing is, you can't drive it, and it hides 90% of the complexity.

To me the value of LLMs is not so much in the code they write. They're usually to verbose, start building weird things when you don't constantly micromanage them.

But you can ask very broad questions, iteratively refine the answer, critique what you don't like. They're good as a sounding board.

carlmr··on Yawning has an unexpected influence on the fluid inside your brain
I yawned when I read your comment.
← PreviousPage 2 of 34Next →