HNHacker News
TopNewBestAskShowJobs

realusername

9,132 karma · joined August 4, 2014

email: hn [AT] alex-min.fr
submissionscomments
realusername··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
Well, both Sol and Haiku aren't frontier model releases, I'd even argue that they only exist because of the competition
realusername··on Yes, and
Most of the software which was the same thing has been abstracted already in CRM, WordPress, content builders, website generators...

What's left is a lot of very different software which have very few things in common.

realusername··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
Maybe it's not intentional but having two providers only making comparison between each other is still a bit fishy to me in such a competitive space.

> For example, Anthropic putting K3 in its comparisons would be a huge admission that K3 is worth considering.

They do put OpenAI though, if it would be a comparison only with their own models, why not I get it but a single other competitor?

It would be like Apple making comparison page with their new iPhone and only mentioning Samsung and nobody else for example, it would sound weird. You either include competitors or you do not and include none of them.

realusername··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
If distillation was as easy as that, we would have 100s of competitors around the world.

China is following because they have an army of PHDs in data science and mathematics and capital to make use of them.

realusername··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
> They also can't coordinate with each other, because that's illegal.

On the last OpenAI release, they only had comparison with Anthropic models, on the last Anthropic release, they only had comparison with the OpenAI models, there's something obvious going on here.

Sources:

https://openai.com/index/introducing-gpt-6-1-sol/

https://www.anthropic.com/claude-haiku-5-5

realusername··on RuneScape's Position on Gen AI
I still don't see any difference with both, there's no product shortcut, in both cases the LLM makes choices on your behalf. On the coding side, it will pick what is the most likely on the parts you haven't specified (remember that any sufficiently detailed spec is just code) and on the modeling side, it will also pick what is the most likely from the parts you haven't specified.

I'm going to pick a real example to illustrate it. Let's say you want to code a 3d sea with waves and everything, there's 1000 ways to do it with every different style you can imagine. If you just say to the LLM "make me a 3d sea", the LLM will make a lot of choices on your behalf and it will influence the art of the game. Even if you have an art director writing a full page on how it should look like, there's STILL missing details that you'd have to figure out on the coding side (and thus, will be decided by the LLM instead of you if you used one)

Maybe if you did it yourself, you would have picked a different solution, different maths, different shaders and the result would be nothing like the LLM has made.

realusername··on Pixel 11 doesn't yet meet the GrapheneOS security standards and may be skipped
They probably know already, it's not the first time it happens. See the illegal pressure Google has made against OEM after CyanogenMod wanted to go commercial.

Google was found guilty by the EU antitrust investigation and payed a 4 billions euros fine (and Google has changed nothing since then, they only increased the pressure).

GrapheneOS is especially annoying for them as it destroys their blanket excuse that it's ""for security""

realusername··on Pixel 11 doesn't yet meet the GrapheneOS security standards and may be skipped
I think cutting it for saving money is the most charitable explanation you can give, the other ones would look much worse for Google.
realusername··on RuneScape's Position on Gen AI
I'd say the same about the assets then personally? They just interact with the output (3d mesh) and not with the process (modeling).
realusername··on RuneScape's Position on Gen AI
> The difference is that game code isn't something that the player sees or needs to care about

I don't think that's true, I could mention a few examples, the cliché of the Unreal Engine hair style or lights and on the other side of the spectrum, the custom made engine which made TinyGlade what it is.

Game development iterates with what's brought from the coding side.

realusername··on RuneScape's Position on Gen AI
Game code is also art, it defines the game as much as the assets. I don't know why the game industry has such a contempt for the coding side.
realusername··on LeCun has "zero concerns" about AI wiping out humanity, recent "rogue" incidents
Why chatgpt is still struggling very hard with photo editing and proportions though? It can't modify anything in a picture without messing the 3d space.

Isn't that a lack of spacial reasoning?

realusername··on Nvidia wants to put a watchdog chip next to every AI agent
AI also had a negative impact on the CI and on time spent to review code & documents so because of that, we're also shipping less
realusername··on Nvidia wants to put a watchdog chip next to every AI agent
> Productivity gains are still enormous compared to what we used to do before agents.

My own productivity yes but if I step back and look at a company scale, the productivity gains has been negative for our company as a data point.

We are now shipping less and with a lower quality.

realusername··on There are no "rogue" AI agents
Same thing, they are aware of the risk and decided that they are okay with it
realusername··on There are no "rogue" AI agents
> but you'd have a hard time arguing that they were intending on that to happen.

The first time maybe, after it happened repeatedly though... I think the opposite, you would have a hard time arguing that they were not intending it to happen.

realusername··on Jev Based Code Review
I can confirm that our code quality at work is regressing and we are shipping less product features than before with AI.
realusername··on Toyota is taking the Corolla electric
Personally I won't, the amount of greed and contempt for customers in the EU automotive world is absolutely insane, they need all to feel the pain, at least a bit to remember that their place isn't a given, that they have to work for customers.

A good chunk of them don't even follow EU laws on warranty.

realusername··on Two-tier encryption in the UK
The only thing where they tried to push back very hard was the EU's DMA, beyond that they folded quickly to absolutely everything else.
realusername··on Ideas on modernizing the open-source desktop
They only superficially look similar but have nothing to do with Windows or MacOS on the feeling
realusername··on Italian parliament votes for return to nuclear energy
I think the complete opposite, we are going towards a high energy world, I expect the current electric grid to be multipled by 3, both to replace petrol and add new usage on top.

Which makes the whole "we'll build renewables and storage" sounds even more unrealistic

realusername··on Italian parliament votes for return to nuclear energy
I didn't downvote but it's nonsense, nobody can sustain a winter with batteries and other storage.

I only checked the first paper and it doesn't seem talk about what would happen if a 3 weeks bad wind event would happen in winter, and a rough look at the figures makes me think it just wouldn't work at all. I didn't check the other papers.

And the proof is in the pudding anyways, if it's that cheap and feasible to do, well let's see who's doing it.

realusername··on Claude Opus 5.5 Intelligence, Performance and Price Analysis (Max)
Personally I don't believe in agentic workflow. I don't think that's a coincidence that both OpenAI and Anthropic chose math problems to test their long agentic workflows, they are well defined, with a clear finish line and with a 100% clear progress path, most of real life tech projects aren't like that.

No matter how clever the model is, most problems have multiple valid, invalid and unclear decisions to make, running it for a long time is just picking the first option on everything, which isn't usually what you want

realusername··on Claude Opus 5.5 Intelligence, Performance and Price Analysis (Max)
I always work on a tight loop, I don't think it makes sense to let agents run for hours.

Regardless of the model, running it for hours means that the model will takes decisions and assumptions alone instead of you.

realusername··on Apple has added persistent 'ads' to iOS, and it's driving users crazy
> But Samsung phones are full of ads for their own and other's services

I have a Samsung and I've never seen such ads.

The kind of ads they are talking there in the article would be in cheap Chinese phones in the Android world I think.

realusername··on Claude Opus 5.5 Intelligence, Performance and Price Analysis (Max)
Personally I use everything in low reasoning. Maybe I'm wrong but I think that the higher reasoning settings are almost never worth it, it's marginal gains for a much higher budget.

I also switch to a better model for more complex tasks, also in low settings

realusername··on I said no and Apple said yes
Well yes it's translated by amateurs but as for the result, both KDE and Gnome are miles ahead in terms of translation quality compared to MacOS or Windows.
realusername··on I said no and Apple said yes
They are translated on your average Linux distribution though, you won't see nautilus but "Files", gnome-software but "Software", Gnome snapshot but "Camera" (I've put the English term here but each term will be translated to your language). See https://apps.gnome.org and switch the language to see.

In comparison, both MacOS and Windows look more amateur with some leftovers not translated. (Like "Windows Update" on Windows for example)

realusername··on I said no and Apple said yes
To each their own, I find the UX absolutely horrible and not intuitive, it's also poorly translated for non-english users (like "Finder", "iCloud Drive" or "Spotlight" not translated in the interface for example).

I have to agree on the hardware quality though but that doesn't do everything as I still have to use the OS

realusername··on In September, AI generated code has made up 17.25% of all Linux Kernel patches
The Linux kernel is essentially the nightmare scenario for a LLM.

No tests, not much documentation to work with, a very high acceptability bar, multiple ways to do the smallest change and every single word on your commit messages is very important.

You can start with a LLM but there going to be some manual modifications you will have to make, even with the best model available

← PreviousPage 3 of 34Next →