jrumbut

The whole AI issue is and always has been a social question. It's hardly a ridiculous concept to have a skilled worker control a $10k budget to buy tools. In many fields that is normal.

Previously, the tools in our field were free or trivial cost. Seems like that has changed, that's OK, there are models to solve that.

I teach a couple courses currently, there is an enormous gap between what my students can create with AI and what I can create with AI. I don't think it's that much smaller than the difference of what we can create without, the only difference is that the schedules are compressed.

The solution is all of the above and more.

Every big problem needs to be addressed from multiple angles, whatever you can do, you can contribute to the solution.

I think the author undersells the significance of this. I could easily picture someone writing code that boils down to the example, all it requires is a flood of signals and a print statement.

If your program generates signals, it could generate a flood unintentionally. If you put a print statement in the handler, you can get this result.

It's not terribly surprising, but good to know.

I am genuinely confused. Are they telling me this because they expect me to be reassured that this anonymous organization is not using my prompts or are they saying "don't expect this particular model to improve as you use it?"

> I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question.

Well, he did diagnose the situation correctly. He couldn't comprehend the confusion of ideas that provoked the question.

I'm also not entirely sure it's an odd question to ask. To this day, users are surprised when their software produces garbage output instead of failing. Perhaps the members of parliament were expecting some form of input validation or sanity checking out output.

What's most bothersome is there is work for them to do.

How about assigning a real copy editor with subject matter expertise? How about publishing open source libraries that automatically validate and output visualizations for their formats? How about hosting multimedia supplements?

It would not be difficult at all to earn the money they charge. There is so much room for creativity and innovation and adding value in scientific publishing.

They used it for both and as a source of vitamin A and whatever other nutrients are in it. They used it for everything. There is an episode of I Love Lucy where a recipe for homemade baby formula is described which includes cod liver oil.

In a lot of things it was pretty good at what they used it for though, that does continue to be true.

It could be my use cases, which have always seemed to be outside the wheelhouse of these models, but I find it very hard to downgrade after accessing a more capable model.

Opus 4.8 produces output in 15 minutes that is 3-4 hours of my work away from output that used to take me 40ish hours (a solid week of dedicated effort).

Last year(-ish, maybe it was 18 months, I forget when the jump happened), the frontier models couldn't touch this work. The output looked like a hardworking intern on their first day. Nice formatting, decent volume of words, but no understanding.

So it might work if it turns out to be a substantial leap in capability.

Yeah, I think the missing piece on this is that the first thought they had was "we can do the same with less" instead of the growth mindset that made me interested in technology in the first place.

And it's amazing they didn't, because most of the tech industry only gets paid in a world where there are offices (either physical or virtual) full of people with money to spend during and after work.

It's still very rare for anyone to be asking "how do we do more with more?" But the person who figures that out is going to be the winner (and if no one figures it out we will all lose, even if you manage to transition to a job that still exists the world around you will be a nightmare).

If you've ever been to the beach, you can smell the salt air and rotting seaweed and hear the birds.

It's all gonna get on the glass (from above and below), and eventually the salt left behind is going to build up. The salt left behind is very hard on any structure or machinery used to move it which makes repairing the large glass enclosure a pain. All this for a slow trickle of water is generally not worth it.

If you have triggers I don't see why you wouldn't put them in a migration. That addresses one of the most problematic aspects of triggers (invisibility, no version tracking, etc) without reducing their usefulness.

With some cleverness you could even introduce some testing that way. Not perfect but better than nothing.

> identify and weed out the worse teachers

By and large, everyone knows.

Data might be useful to tell you "hey that longtime great teacher approaching retirement has checked out early" or "the new hire who was struggling last semester has turned the corner" but it's no secret in a school building which teacher everyone hates and which one everyone loves.

If you woke up tomorrow and discovered you were an elementary school principal, you would have the lay of the land by week two at the latest.

The problem is not separating the flowers from the weeds, it's what will happen if you pull the weeds. Who's gonna take care of that room full of 8 year olds tomorrow? And for the next several years? If a weed shows up every day and doesn't commit any crimes, the downside of replacing them is larger than the upside.

Part of this is the changes wrought by the Internet already. At one point, almost nobody got into fly fishing out of an idiopathic urge to capture little trouts.

I got into fishing because my neighbor liked to take his kids out and I came with. Then I ran into an old man on a lake who could do all sorts of wild casting techniques (through fly fishing) and who explained to me his scientific approach to catching fish. It sounded very interesting when he spoke about it.

The way of sharing information has been upgraded, but the way of forming communities has not. The people who want to catch trout are very well served by modern tools, but the people who wanted an occasion to talk to others in a quiet outdoor space are not.

We need to invent new reasons to be together.

I love people, I love spending time with them. Even though I am married, a parent, and living near to several relatives I still get lonely because of a lack of some forms of interaction.

At the same time, the form of interaction I'm missing is not "debating which font to use on a t shirt." I'm glad a robot can do that for me.

We need some genuine human creativity (or hell, use an AI if it gives you a good answer) for ways to get people to interact in joyful ways rather than over shared drudgery.

Let's go running together and let the computer make a t shirt to commemorate it.

The boundary on this is kind of fuzzy. You obviously wouldn't donate if 100% of it was stolen, but also if you wait until the world is in a perfected state before helping anyone you'll never help anyone.

And for anyone who says "hey that looks familiar, but I've never been to Cincinnati" it was the inspiration for Justice League headquarters.

Yeah there has been a weird belief here for a long time that if something bad happens to us then extremely generous welfare benefits will materialize.

Ask the people who used to work in the auto plants if that's how it goes.

No one will get a dime unless they organize and fight for it. Otherwise things are more likely to go in the other direction, what safety net exists now gets reduced.

Without knowing more about their methodology, it seems like a lot of the recent improvements have involved the AI itself taking time to complete the task.

At first the models turned a 5 minute task into a 5 second task (by 5 seconds I mean a very short amount of time, not precisely 5 seconds). Then they turned a 15 minute task into a 5 second task.

Opus 4.6 completes 8 hour tasks all the time but (at least in my experience) it isn't spitting the answer out in 5 seconds anymore. It's using chain of thought and tools and the time to completion is measured in minutes or maybe hours.

In my experiments with local LLMs, a substantial part of the gap between frontier and local (for everyday use) is in tooling and infrastructure.

That is why I am sympathetic to the idea we are leveling off. But to bring in the air speed example from the article, I don't think we've reached the equivalent of the ramjet yet. I suspect in the coming years there will be new architectures, new hardware, and new ways to get even more capable models.

It's also by far the best in my experience at a request like "it's 3:55 and I need a few slides on the topic of the Gettysburg Address for a 4PM meeting."

I wish it was more integrated into PowerPoint but it's still the best slide generator I've used.

I've lost access to plenty of Claude stuff without canceling anything. I am careful not to leave anything important in there and back up regularly.

It's funny because sometimes it will remember stuff that is lost and not be able to reference stuff that is clearly visible.

One area where I find ChatGPT superior (and this is just my own experience) is not losing things and also respecting project boundaries. Claude projects just seem to be a way to lose things faster, the model seems to be entirely unaware of projects as a concept.

I think it's possible that this idea would work as a communication/branding strategy for senior developers, though I don't think it's strictly true.

I am really skeptical of arguments based around "I can do things the model can't" because that space of things is not very large and is getting smaller every day.

The opportunity to not merely cling on to what we have another year but to grow is to say "together, the model can manage so much more complexity than before that we can do things that were not previously possible."

We haven't identified too many of those things yet, but I am certain they are coming.

Oh yeah, I'm sure that $300 microscope comes in handy but a cheap loupe gets you started. I've found all sorts of other uses for it too.

Or there's an efficient market and when you mail yourself 9,999 $1 bills and take them to the bank you find out 30% are counterfeit.

My typical practice is to write a reply using my own brain and whatever practices are called for, then attach any interesting chatbot responses that were generated as documents.

So there's a clear separation, a reply from me which I stand by and then some interesting chatbot stuff if you're into that.

My father was a big HP calculator fan. I used to forget my TI calculator for class and he would lend it to me.

I was never more unpopular at school than the day we had an exam and I was learning RPN on a calculator that beeped every time you hit the wrong key.

The past couple of years have been chaotic and fearful. Hopefully that won't last forever.

If we can get a little stability, people will begin thinking less in terms of "how do we do the same thing cheaper" and more in terms of "how do we do new things."

I agree with this sentiment but I think LLMs are really close to the Brooks idea of a silver bullet.

I don't know if, overall, it's a 10x improvement or 6x or 14x but it's a serious contender. Part of it is the LLMs are very uneven in their performance across domains. If all I build is simple landing pages, it might be a 100x improvement. If I work on more complex, proprietary work where there aren't great examples in the training data then it might be a 10% improvement (it helps me write better comments or something)

As someone who transitioned from working in startups and technology to a university, it is hard to describe how different the environment is.

It looks very weird and is hard to understand from the outside, and unfortunately all technology vendors are on the outside.

Basically every technology has an impedance mismatch when brought into the university environment. And when you combine them together it keeps getting worse.

That's why you see things in this thread like CS professors who operate their class using pen and paper and maybe a spreadsheet.

I agree entirely. Those of us old enough to have experienced those dreams are naturally going to mourn the loss of the Internet as a place for wild experimentation because we know so much good came from it and there isn't any true replacement.

But it's become clear that in the absence of governance, standards of behavior, and rules both explicit and implicit, the Internet has grown toward tyranny and automated exploitation rather than freedom.

We need to set some rules and expectations that people can rely on, otherwise rules will continue to be imposed on us.

The location it chose was laughably inaccurate (and since I'm the kind of person who posts here I know why). Censoring the IP address was a little cheesy, but down at the bottom it gets better.

It knew how much my phone was charged and it made correct inferences about my device. It accurately read my gyroscope, how I interacted with the touch screen, and it demonstrated (not new knowledge to me but probably interesting to the general public) how these things could be used to identify you and also to make inferences about you (if you are sitting, standing, lying down, etc).

It starts slow but it got interesting.