Nah. No they aren’t. Maybe a few did, but there’s still enough assholes spending money on pushing AI and using AI that it fucks up a bunch of society, still.
They have contracts, the article headline could be correct and everybody would likely still be locked in contract.
Ah see, they’re forgetting to tell AI “make no mistakes”
And use best practices. Or it won’t use best practices. Which best practices? Best practices, you luddite, don’t you know the best practices? Which specifically? Best practices!!!
I just don’t understand this, one can make a wall out of lots of cardboard if it’s ventilated and covered from wind and snow and rain and mist and whatever else. What one shouldn’t do is to expect that wall to be as good as a brick wall.
It’s a tool to shoot slop code where humans too repeat slop code. Replace stupid and sizey code with machine-generated same thing, but cheaper.
One would expect a commercial product to not be something main features of which rely upon generated code. At least without proper testing. Eh, OK, now I’ve remembered how I work.
My favorite is “you’re an expert in SQL”, or whatever tech.
- Tell your employees to use AI without enough training.
- Tell them to keep doing it without enough money.
- Profit!
I’ll admit I’m experiencing a degree of pleasure in thinking about how nervous Anthropic and OpenAI must be with their IPOs supposedly coming up.
They’re already billionaires many times over. Hard to feel nervous with that kind of wealth in your pocket
If they weren’t needlessly nervous (by that I mean paranoid), they would not be in that position. Whatever their wealth, those people cannot feel safe.
Idk about that. Musk and Zuck seem like very careless people.
Musk is never happy. He’s either chasing a ketamine high or getting rage baited on twitter. Real daddy issues with him. Zuck is just a hollowed out human.
Based on the current valuations of their companies. They will likely remain 0.1%ers, but it will still hurt.
Place your bets
Place your bets
They gotta pull Google and get into the advertising game. They are sitting on a gold mine of private data and I’m sure they know it.
The crazy thing is all that data is only good online.
I’ve heard (conspiracy) that the biggest offenders (meta/google) create shadow profiles for people who are not online. So its enough your picture or mention of you is online for them to be add it into the web of collectected data.
Their IPOs are how they cash out and walk away. It’s just a matter of “does it crash before or after?”. Yes they’ll be anxious about getting there, but the fact the market is still behind them and they’re this close is very bad for the rest of us.
Looks like the market is now trying to get out from under them.
OpenAI already explicitly postponed their IPO somehow hoping for a better climate to IPO in…
A certainly… optimistic expectation.
You should not try to make sense of a decision they probably used chatGPT to make. It’s bad for your metal health
NASDAQ just crested a new high this month, before stumbling last week.
If the death of the tech industry is on the horizon, I can’t see it.
DJIA crested a new high September 3rd 1929 before stumbling 3% the next week.
If death of the market was on the horizon, they couldn’t see it.
The slide being attributable to some "I told you so"s coming to pass may bode ill for the bigger "told you so"s that come due in a few months.
But the market doesn’t act rationally, both going up and going down, so who knows.
DJIA crested a new high September 3rd 1929 before stumbling 3% the next week.
And over a century later, what has become of it?
If someone bought $100 of stock in 1929, how much would it be worth today?
$100 of Krueger and Toll? $0
$100 of the entire DJIA on September 3, 1929 adjusted annually: $13,633
If you’d invested in January of 1929: $16,913
If you’d invested in January of 1920: $72,275
January of 1930: $31,708
It’s never gone to $0
This has strayed quite a bit from the assertion that there can’t be enduring problems in the current crop of AI investment because it was at a high a while back.
But to continue, if you invested in the makeup of the DJIA in 1929 and left it in those companies, you would not be having a particularly good time.
If you kept refreshing the investment in the companies deemed to be the ‘good’ ones over time, then yes, your investment fared well, but that says nothing about the enduring value of the companies that made up the index in 1929. This is the current topic. It’s also wild to be this dismissive of the Great Depression as nothing to worry about because, eventually, we decided on a new set of lines that could go up.
So “line goes up” by definition is not really an endorsement that all companies lines go up, we shift the criteria over time to make sure that we measure whatever works in that moment.
IPOs upcoming? You’re being facetious, right?
Guess I need to move some money anyways.
What’s stopping people from using open sourced trained models? Like, why are tokens going to be needed in the future? Isn’t this their whole business model?
You need 200GB of video memory :)
Honestly, I have a little tin-foil-hat theory that the used their infine supply of venture to fuck the memory market to make sure people couldn’t run giant chinese models on solar and eat their lunch.
Of course, why do you think they are running around with warnings about “rogue AI destroying humanity”?
They hope something of what they throw at the wall sticks and some legislations limit their competition.
By the way, this kind of throwing money at the wall unfortunately works (if you have enough money).
But also - with this kind of applications being useful, memory prices would jump regardless.
OK, I’m just guessing.
you mean like Sam Altman’s ‘dirty DRAM’ deal where he secretly negotiated with Samsung and Hynix to (together) buy 40% of the world’s memory production, announced both deals the same day (neither company knew about the other), and bought the RAM not as sticks or even chips but as finished wafers, useless until they are sliced, binned, and packaged into chip casings, a task OpenAI has neither the equipment nor the expertise to accomplish?
For enterprise, I think same reasons why companies using paid software instead of open software. Someone else to maintain and blamed if something doesn’t work.
training a top-level model takes harvesting petabytes / exabytes of revelant data and then months of continuous 100% power on a large server farm costing many millions of dollars.
That was my understanding. So how do local models work? (Been fucking with computers since the 80s, AI is all new to me.)
It’s a one-and-done process, then the normal Q&A usage has vastly lower requirements.
Local models generally means you download the result of all that , after it’s done, (the PT in ChatGPT stands for pre-trained) and then generate answers locally, but the training wasn’t done locally. The generating answers step is called “inference”.
Hardware cost to do inference with a top-level model is within reach of most small businesses. Hardware cost to do training of a top-level model is orders of magnitude higher.
So, there are open-source models that are trained by universities, and then used by affluent but not billionaire people at home.
There are open-source models that can do inference (answer questions) on a typical 8GB gaming graphics card, but they aren’t in the same order of magnitude of quality as the famous top-level models.
Thank you! That’s the best answer I’ve seen on any platform!
What would you recommend as a local model? Want to dip my toes in but I’m woefully ignorant. And again, I’ve done my time in IT, so you don’t have to dumb it down, uh, too much.
The last time I ran one locally was a few months ago, so this might be somewhat out of date, but I’ll say Qwen.
Generally there are two numbers to look at first, the number of parameters (number of nodes + much larger number of connections between nodes) and the quantisation (number of bits that define each parameter)
So an 8 billion parameter model at 8-bit quantisation will require 8GB of working RAM, probably VRAM for fast results at home (at 8 bits to the Byte)
and also a 16 billion parameter model at 4 bits quantisation will also fill the same 8GB.
in my experience, high bit depth enables the thing to respond in more complex terms, and more parameters means more “knowledge” is there to be accessed. I have better results from phrasing questions that’ll have short answers and leaning more towards more parameters than bit depth.
The next number to consider is “context window”, which is basically how far back into your chat the thing can “remember” and have previously given information influence the next answer.
A year to two ago, when I was trying a lot of LLMs for coding tips, Google’s seemed the best for larger projects, simply because it had a much bigger context window than any other, and so could “consider” longer and more interconnected scripts as a whole, even if perhaps Claude was “smarter” with a small problem. I don’t know if this is still true in autumn 2026.
They’ve (commercial models) started keeping a per-account or per-chat log of important facts that exists alongside the context window, but doesn’t carry as much detail. Good for seeming to get to know you, not as good for writing new code that won’t conflict with a long-established bit of old code.
It’s not really practical to have million+ token context windows on home/gaming hardware.
(usually a token is a word, or a couple of tokens are a long or rare word, or a token is a punctuation mark)
LM Studio is good software with generally sensible defaults for downloading pre-trained models and running them on your own PC.
They’re the output of all that data centre processing. The model is fixed at that point and can’t “learn” anything more, so the pressure is always on to release a newer model that resulted from more data with better algorithms, but still small enough to be able to run on a consumer grade GPU
The hardware requirements are substantial. If companies or individuals want to run models locally, especially at the speed you’d expect from the big name models, then everyone needs a pretty serious GPU on their work station. That or the company needs to build and maintain a server capable of running whatever loads they expect to see.
None of this is impossible of course. But you pay for the tokens one way or another. Either you pay the upfront cost of the hardware and continuing cost of the electricity to run it, or you pay with the inconvenience of running a model very slowly on lower end hardware, or you pay a subscription for someone else to worry about all that.
This is why local models are still niche. Only hobbyists nerds are willing to go through the trouble of setting up and running a local model. Most companies still are not sure if AI is actually worth it for them. So it’s much easier to pay for tokens than the giant upfront cost required to build you own infrastructure.
If the numbers the AI companies are putting up as far as infrastructure cost vs revenue are anything to go by, it probably isn’t worth it to most companies.
They thought they sold tokens, they actually sell marginally (6 months lead) improved models.
They don’t need to, OpenAI and Anthropic are running at a huge loss to keep it cheaper to use them than to buy the comparable hardware required to replace them. The only way running open source models is cheaper is if you ignore hardware costs, or if you have a very specific and unusual usage pattern (i.e. constant 100% capacity, and not a clock cycle more or less), or if you go with another cloud provider who are also running at a loss.
If you want it well rounded I can see the argument for enterprise but for me personally I programmed one on my workstation and it’s trained on very specific things I think it barely uses up all 24gb vram I have available.
come on, I’ve trained toy size models too, you can’t seriously say they’re even vaguely approaching the capability of models that takes a whole massive server farm weeks or months to train.
If you do one simple task like I do in my job yes you can make a small one purpose built. Work in reading comprehension before commenting. If you re read which I know might be hard for you, you can see I said make it for a specific task. Example for you since you’re going to try and counter because you’re an idiot. I have to fill in the same multiple fields on an excel table for state documents. This is pulled from a sql database I literally have a model where it does that for me with natural language processing instead of me being a tool and using python exclusively when I already did that to make the easy simple tool. I suggest Amazon for reading comprehension before commenting books. If you message back I’ll send you some recommendations I tell to my students.
Right, and what proportion of enterprise use cases genuinely need an appreciable fraction of that power (because they are wide ranging and non-repeating)?
Anyone that says these things aren’t impressive is kidding themselves. But how do we get from impressive to productivity enhancing in excess of cost? I find smaller specialised models more plausible on that front, but jury is still out on the financials, we’ll find out I guess!
Currently the market allows me to run the AI then do something else meanwhile, and still counting 8 hours work, and meanwhile delivering more than before with less work. It’s my personal Claude subscription and it doesn’t cost much. As long as it works and worth it I will use it.
There are features in the app where code quality doesn’t matter, like a status page in admin panel. I can regenerate any time with a good prompt.
And sometimes there are stuff where I need to use my coding experience, because it’s too unique feature. But still I’m faster, not just feeling faster.
You’re a coward and a fool
Proof you’re a coward: you use Claude to do your work instead of just not doing work and still getting paid. Most corporate places you only have to look busy, not actually to do any work.
Proof you’re a fool: you’re willing to spend your own money to help someone else steal your labour
No I get money while working less. You are the fool. The change is inevitable, the economy needs less and less real programmers. I cannot influence it, it will happen no matter how you hate it. And meanwhile I’m working on the B plan.
And I’m self-employed it’s not a big corporate space.
Ah, so just a fool then
It’s my personal Claude subscription
If your place of work doesn’t already have a security policy preventing personal AI subscriptions from being used on corporate data they probably should.
That being said, many companies are happy to take the additional productivity without having to pay an additional license fee. I suspect this will continue until the prices of the subscriptions rise (or the tokens they offer shrink) to where people are no longer willing to pay out of pocket for the boost. That or companies will realize they’ve been leaking tons of confidential data to these providers.
The customer knows that we use it, in fact it’s a pressure for them to use it, even in the product to find investors these days.
I’d love more of the msm to pick up these articles.
I need to convince my family this current state of AI is just smoke, mirrors and debt, but i need a normie-credible source to anchor it.
Does anyone know of one? Guardian would do.
MSM is too scared to talk about these topics. If the AI companies fall, the global markets will crash. Banks lend out too much money to those companies, same for retirement funds and the owners of those media.
This might be a case of too big to fail and “look out China will use it to destroy us”. That’s what those AI companies are betting on at this point. Their last hope, apart from continuing the deception for as long as possible.
China is already eating our lunch in everything else. At this point, I’m just gonna learn some mandarin and wear a jade necklace.
Why do you need to convince them?
I don’t want them to plan their lives around it. My brother is thinking of switching career and my parents have some investments etc based around line-go-up.
But also I just look like a conspiracy guy until the bubble pops, but then im the “i told you so guy” and that sucks.
I would just tell them to think long term instead of focusing on the new shiny thing. Software development is still a perfectly valid skill to learn, and if your investments are in tech ETFs or mutual funds, they’ll be managed and hedged against avoid losing a significant amount of value.
If they still want to chase the shiny thing, you can’t logic them out of an illogical position. All you can do is recommend they have a fallback cushion, and only invest time/money/effort that they can afford to lose.
They’re not chasing the shiny thing. Theu just think it’s business as usual.
I’m not sure what they’d do differently, but they’d be able to make the decisions based on all the info, as opposed to what they have now.
“hedged against… Losing a significant amount of value” boy do I have bad news for you about the ETF industry.
January: We’re going AI native, please use AI as much as possible.
March: Everyone needs to attend workshops on how to maximize AI productivity.
July: Some models are only available for special (properly justified) cases. Speak to your manager. Turns out that some people did everything (even commits) with top cost models.
August: Ok, we’re setting some limits on token usage, due to some people spending 1000% over the average.
September: we are creating additional frameworks and limits to AI use.
It took a bit of time for the actual AI cost/benefit to reach the top management and put some hard numbers in contrast with the hype machine, but it’s happening.
Everyone who was paying attention even a little bit has known this would be the trajectory from the beginning. And we haven’t even gotten to the crazy price hikes per token that are coming.
And we haven’t even gotten to the crazy price hikes per token that are coming.
Why do you think that?
IMHO, API prices are likely to drop over time as we scale up and efficiencies are found, combined with general pricing pressure from the competition… SOTA models will continue to have SOTA prices, of course, but reasonable priced distilled and open weights models are already common.
A consequence of API prices dropping is that “heavily discounted” consumer plans will also adjust and won’t be so generous relative to API prices.
Because these companies started selling at a loss from the start to get people hooked. Neither Claude nor OpenAI are turning a profit right now.
The biggest problem isn’t cost of running the models right now, the biggest problem is all the debt they built up on training and data center buildouts. That’s what they need to recoup and that’s what’s going to drive prices up. And once it’s recouped they’ll need to make pure profits.
And I hope that debt kills them because all the big AI companies are terrible in one way or another, and the eventual price disparity will make it uneconomical to use their models. Then we might return right back to development focusing on small models, not large ones.
But aren’t the AI companies always needing to train the next version and replace/upgrade the HW? If it were just a one time startup cost then maybe they could become profitable, but if they will always need to invest heavily in traing the next gen and replacing HW every 2 years, then that’s more of an ongoing expense.
And I’m sure you’ve noticed how all the revenue and profit numbers that leak out ignore “one off” costs like training and data centre building.
But look how dangerous these models are. They need to be regulated so that only responsible closed model providers can produce them.
Just ignore that the responsible closed model providers are the ones opening their sandboxes and training their models on blackhat sources.
IMHO, API prices are likely to drop over time as we scale up and efficiencies are found, combined with general pricing pressure from the competition… SOTA models will continue to have SOTA prices, of course, but reasonable priced distilled and open weights models are already common.
Some of this seems like wishful thinking to me.
“as we scale up” We already have large data centers, and, if anything, “scaling up” is just driving up expenses for debt and hardware. The real problem with AI, relative to the internet boom, is that the more you use it, the more expensive it is. The economies of scale to be taken advantage of are minimal.
I do think price of compute will come down, but exactly because people will be running their own open source models on their own hardware instead of Sam Altman to stack GPUs in a warehouse, but it will still be too expensive and inaccurate.
And we’re still headed for a financial cliff in this scenario.
Like a fraction of a percent of people will run their own models. It won’t influence price. Remember that coders/tech people are only a minority of users.
Oh, I see. I was mistaken. I thought tech companies were OpenAI’s major customers, and they certainly could make their own data centers for internal work, and that coders–that minority of users–were the only ones likely to pay full price for tokens, the rest of us wouldn´t be willing to pay full fare for a machine that is right most of the time sort of. Thanks for setting me straight on that.
Some large companies do it … but I think you greatly overestimate the willingness of large enterprises to build and maintain their own infra. Once a company has lost the competency (and many have), it’s hard to get it back. There are always pockets of people with the right skillset in a large org, but they mostly don’t want to do DC work anymore. Team/personal servers and beefy laptops, otoh…
Of course they want to turn the corner and ideally keep all their infra, but they might need to restructure their debts if/when the creditors come knocking.
BUT: they have a money printing machine with inference… Enterprise is where most of the money is made, as they pay (closer to) API prices… consumer & small business plans are heavily discounted in comparison, with mostly underutilized use-it-or-lose it plans. They’d be stupid if they were making an opex loss on inference.
As for why API prices are dropping? Competition is a bitch… but they seem willing and able to play the game for now.
If we’re talking about the survivability of AI as a technology, I agree.
If you’re talking about the viability of the hyperscaling business model, I think we are looking at wildly different sets of information.
They were supposed to do their IPO and have positive cash flow already. Time is running out before investors are getting antsy and there’s no more money available for scaling up.
They gobbled up the easy money… now they have to work harder for more, and yeah, they may have to slow down a bit and embrace fiscal responsibility… I expect to see restructuring of debts along the way, but I’d be very surprised to see any of the major US based players implode.
Both Anthropic & OpenAI have debt repayments coming up in 2027 that:
- cannot be restructured
- cannot be postponed
- they absolutely cannot pay back without even more VC money, and that faucet is being cut off right now.
Both Anthropic & OpenAI have debt repayments coming up in 2027 that:
- cannot be restructured
- cannot be postponed
- they absolutely cannot pay back without even more VC money, and that faucet is being cut off right now.
Well then. I stand corrected. The random Internet person with unprecedented access into the books & private financial dealings of two major competitors has an air-tight argument.
I suggest we come back to this thread in a year’s time and see how it is playing out.
Look up some well known critics in the AI space and you will see a bunch of information that aligns with OPs statements. Its always been scheduled for Q2/Q3 2027.
Keep spending that shit left and right!
I feel like we’re co-workers, with how perfectly that timeline matched the day-job’s progression.
We’re at “did you take the seminar on looping that uses tokens like a firehouse so you can use less tokens? Why are you using too little (/not enough) tokens, and why is it not magically 10x-ing all of your (fucking ansible) config?” Or some shit.
Do we work at the same company??
I think this was the standard management reaction across all the companies that embraced AI as if it was the second coming of Jesus Christ.
C-Suit are herd animals. Same schools, same country clubs, same conference. They all are formed the same. And they don’t take advise from people who earn a few orders of magnitude less.
Fortunately, the people above them care about numbers, and these don’t add up.
They thought it was their ticket to fire everybody, and the more they made you use it the sooner they could fire you.
Soon, the guys upstairs will realise that the real savings are in replacing the middle management with agents, not replacing the workers. Fingers crossed.
That will just become a different kind of hell
I don’t doubt that for a second, my friend.
I was a PM at a major now-hyperscaler and one of the engineers on my team was pretty vocal about PM not contributing.
He ended up leaving for a team that didn’t have one. A few years later I happened to be in his new building, he saw me, and came up with this huge grin on his face, “oh wow are you joining our team?”
You always hate the chick writing the specs until you don’t have one any more. 😂
My company is at the July stage. Who knew giving people unrestricted access to Claude enterpise with no training would blow through budgets in days
Our management is so “forward-looking” that we are still in the March stage: they want AI used everywhere, and added everywhere in the product.
Saying that sometimes you still need to use the brain for big decisions will make some people question your prompting skills publicly.
Hell, we even rebranded the whole company to be an “”“AI company”“”. I hate it there and I sense layoffs coming soon, but finding a job in this period is hard. Wish me luck.
Still in March over here. My management is pushing hard for us to start doing shit in copilot, and I’m humoring them, but I’m looking at current news with a very Ralph Wiggum “I’m in danger” mindset
At our company, 12 months ago they said everyone wasn’t using enough AI. They even made a leaderboard to see who was using the most tokens in the business and encouraged more and more.
A couple of months ago, everyone’s AI spend was restricted to $100/month without warning. Entire workflows broken overnight because people had built theirs around Claude skills that used stupid amounts of tokens.
Now they’re starting layoffs. Go figure.
now they are laying of staff? What, did they blow their whole budget out?
They were burning over $250k every single day. So yeah, gotta pay the reaper I guess.
Can’t wait to see this exact headline another 50 times but none of the companies in question facing any consequences for their actions, profit driven society is so fucking stupid.
The consequence is that they waste money and time. Which actually matters for businesses strapped for money or time. But most of the megacorps are swimming in money and in time. They waste those all the time even without the AI because the upper management accepts kickbacks to contract with the worst (but also the bribiest) contractors. And mid management, and everyone else, put their own personal interests ahead of the company’s. Now that the AI hype is here, they have a new tool to waste time and money with, but nothing has fundamentally changed. Big business is a lumbering inefficient bureaucracy by its nature.
Ernst and Young say 36% of companies using AI have experienced catastrophic incidents.
Hopefully some of them go bankrupt. That would be a stellar way to get rid of some bad management.
Hahaha. No. Management is not losing jobs over this. Not in the quantities needed.
If anyone important does get fired they’ll just get re-hired by another company who needs some doofus to fill a position in upper management.
“And why did you leave your last job?”
“The company went bankrupt due to severe mismanagement.”
“How did you get your management position at that company?”
“My dad was rich.”
“You’re hired.”
“Now watch this drive”
You forgot the part where they’re on the golf course and already know each other and are acting like there’d be an interview because pretending to be poor and like the world is based to any degree on merit is actually hilarious to them.
No I know it’s a mostly futile hope.
get rid of some bad management.
Good luck. Like cockroaches, they’d probably survive a nuclear holocaust.
Is it because they turn bad once they become management? Or were they like that from the beginning?
Personally I think it’s both, but mostly the latter.It’s a feedback loop.
Upper management can’t tell what good work is, and thus whomever is the most shameless in talking themselves up wins, unless they royally screw up in a way that can’t be missed. If you have any shred of humility or nuance about your work, you are almost certainly not management material.
I have met like 3 or 4 people that I actually had respect for in management roles, but overwhelmingly they were there solely on an overinflated sense of self importance rather than substance.
Both. People who are promoted to management because they are good individual contributors are often lacking in management skills (they turn from something they are good at to something they are bad at), and people who crave and hard pursue management positions rarely have management skills (always bad).
That sounds extremely plausible. 👍
Frankly, I think we all have a bad manager inside of us, and the problem is that being put in a manager position exposes you to humanity-draining incentives that feed the bad manager parts of you, until it defines you. Higher compensation and promotions for focusing on costs and not quality, and short-term gains rather than long-term stability.
If you’re unusually self-aware and smart, and put effort into grounding yourself, you may counteract it, but most people probably just float through life and let those incentives define them and their choices.
Interesting point.
It also attracts narcissistic people without empathy (and sadic people), a bit like the church attracts pedophiles. Not everyone is, but in the end it’s too many (over 0).
It also attracts narcissistic people
This has AFAIK actually been documented.
If you have more information about that, I’ll be all ears!
I wrote AFAIK, which means I don’t have a source at hand, only memory of having seen the claim documented by research somewhere…
It is well observed that psychopaths (a.k.a. sociopaths) are found both in prison and in managerial positions.
That comment is made by a PhD psychologistm and confirms it.
Feel free to investigate further, at least you now know there is something for sure.
Ah yeah, bank managers (AFAIK! or heard) is often people without empathy for example, guess some jobs attract some types of people.
Both. Whether you’re considering empathy or competency.
Increased power—in pay or in authority—may be followed by decreased empathy (they turn bad). Conversely, dark triad traits have been linked with success like promotion to management (they were bad already). The research on this side has been criticized, mainly for methodology. I think the underlying idea makes intuitive sense.
On the competency side, there’s the Peter Principle, originally satirical but subsequently researched. Good employees are promoted to a position of incompetence (they turn bad, or rather, are promoted until they are no longer good). Or, they were incompetent and promoted anyway to get them out of the way (they were bad).
And the mechanisms are not mutually exclusive. E.g., a good employee is promoted, they are bad as a manager, then entrench as their new authority leads to less empathy.
I think they get promoted to management because they’re bad.
They’re not going to promote the guy who knows what he’s doing, because they need him to keep working in his current role. The guy who slacks off, bosses people around, and gets chummy with the boss though? Perfect candidate for promotion.
I tried AI-everything (apart from code review of course). But it’s just too hard. Letting the AI do bigger things results in subtle, hard to find, but devastating design fails and errors. It’s fractal. No matter what layer I look at, it’s always sort-of working but also bugged in subtle and sometimes hard-to-fix ways.
So I don’t do that. Instead I use the AI as an assistant while I still do most of the actual work. I downgraded my account too because I never reached the limits for a while.Ding ding ding. It’s fucking GREAT for eliminating the menial bullshit like scaffolding, data cleanup, sifting through overly verbose documentation… Anything that requires abstract thought still needs to be done by a human.
I’m in a similar situation. My workplace has been heavily pushing implanting AI tools into our daily workload, but I’ve struggled to find a more meaningful use beyond a technical search engine: eg. How can I calculate x in the most resource efficient way?
My IDE told me that there is a new version of some library which fixes security bugs. I asked the AI what will break if I update. It actually found out that the new version requires a license key which has to be updated yearly even for the free use case. So I had it refactor the single use of that library to use another library instead. I then polished the refactored code myself a bit.
When I am done with something, I ask the AI what it thinks about the changes. It then tells me where I violated the style guide and sometimes it finds silly bugs I somehow missed. In rare cases it also finds serious stuff. the code definitely never got worse by having the AI look over it. Btw, looking at it doesn’t mean changing it. I had to tell it to save in its memory files that it should only change anything when explicitly asked to.
When I am not sure why something is like it is, I ask the AI. It then comes up with one or more potential explanations and often I then remember what was actually the reasoning if it’s my own code or the code becomes somewhat understandable when it’s written by someone else.
When some behavior looks like it should be impossible, I ask the AI how it’s possible, and it usually tells me the actual reason or gets close enough to make it much easier for me to find it.
When I’m unsure how to implement something, I discuss some options with the AI. It’s reasoning isn’t always sound. But sometimes the internal dialog is just not enough and chatting with AI works surprisingly well for me.
Just treat it like an extremely knowledgeable college with a serious psychedelics addiction. Can’t let it do stuff unattended. But when it isn’t hallucinating, it is pretty capable. Also, it doesn’t care how much you ask it. In general, every easy to explain task with easy to verify results and low damage on failure can be done by AI. But it is often faster to just do something yourself than explain it to the AI and polish the result.
that it should only change anything when explicitly asked to.
Does that work?
Yes. There are the rare exceptions. But since I had to tell it to save to it’s memory that it isn’t allowed to change stuff without a command to do so and that it shall treat my questions as just that (this is likely the actually important part). Then I had to scold it a few times when it ignored the rules. And since then, it actually does obey very consistently.
I use Claude Code, but would expect any frontier LLM to have some “memory” system (in case of Claude Code just an index file and one file per memorized rule or fact).I first saw something like multi-session “memory” in AIDungeon years ago. It was slow and with only a few kB of context not worth using. But that was literally a toy.
Also not the person you responded to but in my experience, yes. I don’t tell it to “save it in its memory files” because I don’t think that’s a thing, but including “do not actually edit any files, only propose changes” somewhere in the initial prompt of a conversation has worked well for me.
Not the person you responded to, but yeah, it seems to in my case.
I’m one of those people that don’t exist, in that any AI code I generate is reviewed prior to running. I’ve been forced to use fucking copilot to check in the source I have to a new company repository, and so far diffing the code checked into source control is unchanged, but it seems to be good about updating the header to the standard in supposed to be using. I’m also having it do basic reviews of the code and only updating the files as I see fit.
This just in, Management are complete morons. Film at 11.
Trend chasing upper class idiots
This. You’d almost think people are dumb enough to blame bad management decisions on software.
If only literally everyone warned them this would happen…
But [company] selling [thing] told me it’s good, it’s not like they would lie about that!
“Why would I deceive you?”
She just exudes “Gollum pretending to be Sméagol” energy.
I feel some very disturbing mental health issues.
The salesman took me out for a steak dinner and then a strip club! He’s my friend!
No no no. Most companies already knew this. Most bosses just pretended it wasn’t happening so they could ride the bubble and bonus themselves… We all said this was what would happen, and it did happen, and only complete suckers would still have fallen for it. Of course some CEOs are suckers, but many of them are simply greedy.
We all said this was what would happen
They only listen to people who’s wage or billable hours are close to the order of magnitude as their own.
You mean malicious compliance works for the mid management too? I believe it.
Someone sent me what was clearly an AI generated doc. I read it, realized it was all AI, had my AI summarize it and ended up with nothing that brought new thinking
What a loop…
This is the thing that drives me absolutely nuts with how most people use LLMs at work. I’m fully capable of asking the LLM to create slop, I need others to actually check and ensure their documentation is useful and correct before communicating it with others.
What, you don’t want to read a 10 page doc made from a 2 sentence prompt?
Just have AI summarize it into 2 sentences
you are part of the problem
They’re just quoting the article… Which you’d know if you bothered to read at all, or even just look at the fact that it was quoted in their comment.
I did read the article, but missed that this was a quote. I stand corrected.
It happens. Props for not dirty deleting.
