lnenad

How is it pushing back, if anything it's mostly clarifying and the conclusion aligns with the original article:

> It’s cool that Astra was able to make direct weight updates. But we shouldn’t think of this like a radically new capability that will change how AI research is done. We should think of this as analogous to the ability of LLMs to emit machine code and tweak compiled binaries: it’s an impressive trick, but the best way to work with binaries is still to write source code and compile it.

It's like people are immediately forbidden to be impressed that AI was able to do this. The original article itself mentions the mechanisms by which Astra achieved this. They just of course wanted to gain more attention so they used a sensational title, like 90% of scientists that want their stuff read.

How is it not moving goalposts, it knew how to detect which weights to update and a way to do it? Even this article says so in the part that I quoted. The original piece never stated humans are unable to do this, I don't even think it implied it. So being impressed with this accomplishment is actually *ok*.

We keep moving the goalposts, I love it. A few years ago LLMs writing simple functions was downplayed and now this. I lovehate the future this will bring.

What a stupid baseless comment with 0 context.

I'm asking primarily as I've seen people mention this as a non fully polished feature on the current array of XR glasses.

How steady is the anchored screen? Does it wiggle?

Building software makes me happy. I understand lying in a cave with a PC for 80 years isn't smart and that's not what I'm suggesting (if your statement implies this). I'm also not saying this is a single variable that guides my decision making process around how I use my time (or how others should). But it's different to how I'd feel if I hated my job and some brogrammer on linkedin suggested grinding is the way to go as money = bit*es.

I mean as with any other blanket statement it's not a single variable, right? But if I feel like my work is torture I'm not gonna give it extra time and effort. If I feel like I'm enjoying myself, then I'm not even looking at it like I'm doing anything extra. But I mean it's just my opinion.

Do you get satisfaction from your work? If yes, then that statement is irrelevant for you. If not then clock out at 5pm and do stuff that brings you joy and satisfaction.

I was actually using powershell and omp.

Yes of course, it's burned into my mind lol. Local GLM5.3-Flash, 4.0bpw running on a tuned exllamav3 server via omp, max effort 256k ctx. It was trying to cleanup after itself by deleting some temporary files it left behind the sweet thing. I look at it like a child so I take the full blame for the situation, and again, it's a learning experience.

I've had a bad situation with a model adding a space between parameters `rm -rf ~/ folder` and nuking a portion of my homedir. This was of course running out of sandbox since I'm a fool, but lesson learned and I immediately implement sandboxes with nono and it's always fun to see the data after the agent session is complete, I've seen some weird things/URLs appearing. I've attributed it mostly to omp telemetry n' stuff but I'll try to be more careful about that as well in the future.

Yeah that tracks completely with my experience as well.

Yeah, but if you need to provide less details to Opus to get the same output it means the model is better at deducing your needs, which again implies it's a higher quality model.

I agree, but my point wasn't about being able to use a weaker model, it's about benchmarks not being in any way representative of the actual quality/level of the model. Of course for a lot of scenarios Opus is overkill and Qwen/DS are more than enough.

> there’s no major difference here

Sure thing, I can see any sort of serious discussion is off the table.

Same. Every time I prompt some other new/best/benchmaxxed model, I think to myself: wow, finally, this is amazing. And then I give the same thing to Opus and the bastard is always better. It's annoying to a point.

It's not just 5.5, every iteration of the frontier moving - Opus has had some edge. I'm sure it's not 100% in every scenario, I love interacting with Codex models more, but the quality of the output, in most scenarios that I've been exposed to, has been the best.

Every single model I've tested that is supposed to be anywhere near Opus, wasn't there. There is some magic sauce in Anthropic models that these benchmarks aren't demonstrating.

> Customers don't care about your software factory, they care about a product that works reliably.

You don't think AI is capable of putting out CRUD apps into the ether that work reliably? Even my local GLM5.3-Flash is capable of doing that. These products make up a non negligible amount of the industry. Additionally reducing headcount is about to hit in the next year or so. Where are juniors gonna find jobs? Etc etc...

Zen turned to shit for me in the last month. So many stupid bugs started happening. For example, clicking on exes in the downloads would retrigger the download instead of running the file. Random crashes. I was able to drag the whole window by clicking on random parts of the page (I noticed this as I highlight text during reading). It got very slow. I tried to report these but the button to report would actually redirect to Firefox's feedback page lol.

I switched to Brave, migrated in a few minutes literally and feel much better.

If the only variable is money, local models will never make any sense at these prices of infra/subscriptions.

For a lot of people, there are other variables though.

In short we no longer see text based reasoning traces meaning it's easier for the models to fool us in case of malicious plans.

I agree 100%.

I mean that's fair. I disagree as a European though :).

If you take a good SWE salary in Munich (example 80k), and compare it to an average/solid software shop in the states that gives 200k. Expenses can comparatively quadruple and you will still have much more at the end of the month to do with as you please. You can live in so many places in the states where this salary is realistic. But only a handful of places in Europe even have salaries that could come close as a concept. Switzerland that could get you there has even higher expenses and even there only FAANG offers such salaries.

And outside of software, doctors in the states make many multiples compared to European ones. Most blue collar workers make multiples, lawyers make multiples, pretty much any job you take you can make much more. I agree that life is better in Europe and I'd choose to live here 9/10 times, but that's down to mentality and the fact there's a reason we're less productive than you.

How do Dubai, SA, Qatar and similar fit into that picture?

It doesn't scale linearly. You'd be underestimating how expensive it has become in most of the third world now-a-days. You'd be making 10x+ while spending a few times more. Spending also has a cap (depending on lifestyle). If you're in engineering your salary continues to grow with experience.

People are moving to the states not because it's a great country, it's because they're able to make money there. More money than in their home country. This of course is not an indicator of *general* attractiveness but it still makes the country itself attractive as most of the people are seeing more money as a way to have a better life.

This is all good and full of great opinions but these are the same arguments used by artisans during the industrialization against machines. Does it matter really to hand over a large chunk of this work to AI, in the grand scheme of things? Not really. Most python apps were slop written by humans, the analogue of mass market goods. Same principle will apply here, high quality goods are still usually hand made even today. From a software standpoint, if you need something niche, max performance, max security or whatever you will probably hire the best humans instead of handing it over to Opus 7.8.

The wheels are in motion and there's pretty much nothing we can do to stop them.

For slower machines it's much better to use a cloud model, but Qwen3.5 4B can do the job and is fast enough on older machines.

I literally today pushed v0.1 of my dictation app that does both in a single package. You can connect either to a cheap LLM but I've set it up out of the box to use Qwen3.5 9b which does the job well for free, and everything stays local, no telemetry. You can also use Claude/OpenAI/Local providers. https://github.com/lnenad/lipwise