HN user

EagnaIonat

379 karma
Posts2
Comments255
View on HN

Seems obvious when you say it like that,

That's exactly what happens. CSAM has two parts to it.

The first is the known and most horrific videos of CP out there. These aren't shared, instead only hashes are used to determine if the image/video matches. It's next to impossible to get an mis-hit on that.

The second is nudity and age detection AI. Apple has this on their devices, but warns the user before sharing not reporting.

It normally takes more than one hit to get flagged for investigation. At that point law enforcement are informed and take over.

The majority of investigations are children sending naked pictures of themself to their friends.

So it is absolutely being used as the purpose to protect children and it's amazing how many people don't realise how it works when they are very transparent about it.

Of course it is not without its issues that need to fixed. For example providers will just ban the person before it's determined if they broke a law or not.

Interesting (short read) but the premise can be applied to everything.

Automation, Programming module/libraries.

But for replacing others we don't care about, often the doorman fallacy is hit.

They are probably talking about data centres and town hall meetings where people are objecting and being removed.

Even a casual search you will find numerous videos.

The only issue I see with your comment (and this thread stems off the misreading of my post), is that despite your "I see the difference" it is still labelling one side.

In truth, there are crazy people everywhere.

It's just a really good illusion from a well tuned stochastic parrot.

The stochastic parrot claim is long gone.

Someone with actual skill in their respective field can create something new and original from current AI systems.

The real danger is that creative/intelligent people are giving up those abilities and letting the AI do it for them.

It depends, but I from what I have seen on multiple projects with different teams/skills, that it amplifies the ability of the person/team.

It pushes the bulk of the work to review. So teams with good practices can account for this.

For junior teams, the time saved is massive, because they aren't doing all the other practices required to prevent technical debt.

So, E2E is unaffected?

Because its an extension of an existing bill.

After the whole internet lost its mind about Apple CSAM scanning and being horribly off target, I'd recommend to ignore news reports and go to the sources when making an opinion.

https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex%3A...

https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=COM...

The only issues I can see is the error rate for non-CSAM scanning and how the latest vote was conducted.

Although the scanning part, it is still law enforcement that has to review that information and determine if there is a case. Most of the those are teens sharing naked photos with other teens.

Even with the motion/background button toggles you are still left with tiny fonts that make it hard to read.

It actually made me check my browser wasn't set to zoom out. But then using zoom changes nothing, which breaks accessibility.

Also why does the header need to take up 3/4 of the screen?

[edit] Clicking on a paper doesn't even bring the paper up. I have to hit another click to get to it.

I rented a car that had this feature and instantly hated it. I forget the make, but if it detected you were not looking at the road for a period of time, it would slam on the brakes and release. A way to jolt you awake.

The problem was it routinely misclassified where I was looking, so I would get random brake slams which would panic me more.

I returned it after a day and asked for a new car.

Both machines will be stuck in the 30-50GB model range,

That is not true.

3090 is only 24GB.

It will absolutely outperform the M5 on models that will fit in that space, but M5 will outperform with larger models as the whole 128GB is GPU addressable.

With a dedicated GPU, the lag is in transferring data to the GPU. You don't have that lag in ARM.

But it really depends on what it is you want to do. An MLX optimised recent model will run fine and at decent speeds. Granite4.1 (a few months old) for example takes up 2GB of memory, insanely fast and results are good vs much bigger models like gpt-oss-120b (a year old). It even runs on an M1 mac with good speeds.

The models are only getting better.

MacBooks with their unified memory behave like a slow GPU with enormous amount of video RAM. So you can run large smart models slowly.

With the model using MLX the speed increase is night and day. Even non-MLX is good.

You also don't have the transfer costs related to moving CPU data into the GPU.

thers, like coding or more advanced agentic workflows can demand much more powerful models.

You can do coding and agentic fine. For coding I use qwen3.6:35b-mlx and agentic granite4.1:3b works fine.

These are the models I use.

- granite4.1:3b

- granite4.1:30b

- gpt-oss:20b

- gpt-oss:120b (less so now)

- mistral-small3.2

- qwen3.6:35b-mlx

There will always be use cases that don't sit on your laptop, but most of what can be done can be done locally, it just requires a good framework to sit on it.

Depends what you need the model to do. The recent granite4.1:3b just takes 2GB of memory and is fast. Results are pretty good and support tool calling. Barely a squeak out of the Mac laptop.

Even faster with the MLX builds.

Then when I need more heavy lifting I fire up a larger model.

IMHO the issue isn't the models. I've had OpenClaw give the same results as Claude using open models locally. Slower but does the job. Something that can do optimal model switching is what's needed.

This. Here's a quick experiment I did yesterday.

It's like running a sports car and then complaining it burns through petrol too fast.

The truth is the model while impressive is not needed for much of what people need.

Local models can do the work and just offload heavy lifting to the cloud models.