VBprogrammer

I think there would be no point in making the claim if the probability estimates were truly in good faith.

I suspect the easier option is to use AI to review your own code. The arms race between developers and attackers might even make some of the AI valuations come true.

It really feels like we've crossed some kind of event horizon with AI exploding in so many seemingly unrelated directions.

I guess life would be boring if we were all the same. I found the 1.0 litre turbo in the Sandero to be perfectly adequate in power. I hated almost everything else about the car though. And again, I don't usually have this problem, more often I fall in love with the little rental cars we get and start checking how much they are on the used market before remembering we have no need for a second car.

Isn't the fix to constrain the abilities of a user agent to only the permissions of the user inputing the prompt? I guess that's not a lot of fun because you have to implement some kind of query API which respects user permissions on top of the underlying data storage rather than just letting the agent have at it. Any fix at the LLM level seems destined to fail.

Chances are most manufacturers are going to use a cheap USB camera. Can a raspberry pi emulate webcam? Just place the same video of you diligently staring out of the window on repeat.

It seems like you are being downvoted but I've had the exact issue you mention where there is heavy over-banding on the road surface. Or where you try to move out to overtake a cyclist and it decides to correct you back into lane.

Ever driven a Dacia? I had one for a rental in Portugal. Honestly the least comfortable and most irritating vehicle I've ever driven. I'm not just being fussy, we've had plenty of Hyundais, Citroens and the like without a problem.

The snap method of tying a bowline is my favourite. In nice rope it looks really fancy and can be done mostly from muscle memory (except remembering to pass the free end through from the backside to avoid a shameful cowboy bowline).

I found myself in a similar situation taking my kids crab fishing. We were quite high above the water so I was trying to tie the thin string on the net to the bucket so we could lower it into the water to fill it. After trying to think of the fancy way of doing it I just went with doubling up the string and tying a round turn and two half hitches, an embarrassingly simple knot which has most of the advantages of anything more complex.

Knots in general are a harmless and occasionally useful pasttime for many people. In fact I have a length of rope under my desk I will occasionally pick up and practice a few useful knots in just to keep them in my muscle memory. It's similar in many ways to people who solve rubix cubes or play with fidget spinners.

Those are completely orthogonal to accepting all of the risk, from the point of view of a CEO, of battery failures for cars you sold 10 years ago.

But yes, making things repairable is really important. And, I certainly have concerns that the danger of high voltage batteries will soon be used to introduce so much regulatory burden that it becomes impossible for individuals and small companies to repair them.

That's easy to say when you aren't the one considering adding a huge liability to the companies bottom line.

There are also some elements of EV ownership where they require some TLC, like not overcharging them when doing mostly local journeys, not using rapid charging more than necessary etc. If manufacturers were carrying the can it would be easy to let these things go out of the window.

I used to regularly go to a comedy show on the site of an old brewery in Wandsworth. The comedy cost £20 but the beer was free. Apparently the site was the oldest continuous used brewery in Europe and the head brewer decided to keep the record going after the site was sold to a property development company. However, there was a prohibition against competition and licensing issues meaning they couldn't charge for beer.

My partner had booked a table for lunch for us and our friends. Six adults and six children. One of the couples had forgotten a party earlier that morning, so we tried to move the booking a couple of hours later.

Unfortunately the only phone line was answered by an AI bot who stubbornly refused to move the booking, simply telling us there was no availability within an hour of our booking.

Fortunately my partner was passing so was able to go in and speak to someone is person who was happy to move our booking back 2 hours. Lunch and drinks for our party must have come to several hundred pounds.

I'd estimate our party was between a third or maybe half of all the customers there. Had we chosen to book elsewhere I bet someone would still be patting themselves on the back about how clever they were to save a few minutes a day on actually answering the phone to actual customers.

In any ordinary times this would be an incredible endorsement. I'm not sure it works in the current timeline though.

This feels like a stretch. It would be impossible for someone who didn't know how an engine worked to repair or improve the design of it.

It's merely a ToS violation.

> Can there be bugs? Sure. That's the price of not reading or understanding every line.

I've yet to come across a human developer who's output would meet this standard, despite writing every line.

In fact, having an LLM review our code is catching quite a few bugs before it reaches QA.

I have a vague memory of there being some case law or interpretation here. But certainly doesn't bare much resemblance to what the original law makers had in mind.

I've often wondered what basis companies are using for the "opt-in to tracking or pay to opt out model". It has spread now to even fairly reputable organisations.

This, at least to my understanding, runs contrary to the spirit of the GDPR regulations. Permission has to be freely given which, when the alternative is paying a subscription, it quite obviously isn't.

In general humans don't have perfect recall. Even people with what we might call a photographic memory don't have the ability to memorise millions of lines of code and output them with little effort.

It hinges somewhat on the concept of how much you believe things are being learned and how much is just pattern matching and borrowing a solution from memory. Certainly in the early days of Copilot it was possible to get it to output chunks of open source code near verbatim.

I think, generally, people are probably closer to believing that there is some kind of reasoning being carried out by these models than in those early days but it would also be easy to strip all of the immediately identifiable comments etc from the training materials to make it harder to detect.

I wondered how long it would be until someone took the obvious step of wiring a voice assistant up to a full blown LLM model. Seems like the thing I failed to consider was whether that was actually a good idea.

> They'll just be far fewer than they used to be.

I do tend to agree. Though at the current pace of change I don't know if we can take it for granted.

As a recent example, I was on a chat with the two most experienced technical people in our company and the original developer of a feature trying to work out why we were getting a null pointer exception in a very specific case. Of course we had a fix, just a guard against the null pointer, but I'm always uncomfortable with not knowing the underlying cause.

I kept digging while someone promoted the fix. Eventually ruling out two of our original theories as to why it happened. Until eventually someone just asked Cursor which spit out a theory which matched the symptoms perfectly and which we quickly reproduced locally.

I still think we'll need some kind of human who lives in that wide space between the 95% of the population who couldn't get Excel to sum a list of numbers and the machines but the industry will be unrecognisable.

Honestly, with the caliber of people who currently comprise the US administration; leaving the whole thing to Openclaw and some new fancy model might not be the worst idea.

I think there is going to be a long time before all of the obscure knowledge of a decent software developer can be completely replaced by AI. Though the job is going to change beyond recognition. It already has in many ways.

Technical debt is a indefinable quantity which makes it very prone to be abused to mean "I wish I could rewrite this in [insert some fashionable language, framework or coding style]".

AI slop is an easier concept to quantify. It's basically the code for which insufficient people in the organisation have a meaningful understanding of how it works or what it does.

The crocodile tears of companies who've hoovered up everything possible, regardless of permissions or legality, now crying that someone else is stealing their hard work is comical.

I don't even think they can believe it themselves, it's in reality they are just trying to throw fear, uncertainty and doubt about potentially cheaper offerings.

I have my use covered by my employer but we also have budgets and limits.

I think we're all past the "bet-money-can-buy" stage. The most expensive models are an order of magnitude more expensive than the middle ground ones, so you need to be selective about what you run where.

And with a bit of careful routing - there isn't a lot stopping you sending the hard stuff to a cloud model and the average stuff to an on prem model.

I'd think for most companies the pace of change is too high at the moment. Give it a few years, a bit of a plateau in the improvements in frontier models and I can't see how many of these companies don't implode under the weight of competition on inference prices.