102 points larrysalibra 1 hour ago 92 comments
origini 1 hour ago | parent
Which also means that it is slow in adoption not because the quality is not here but because you need to do things first.
You need an env to run an agent, you need connectors, you need to get that stuff through security etc.
The only question is timeline not investment or bubble because AI as is will not go away anymore.
I asked before for headcounts or consultants and today I can just use AI for it. Its not the AI which is the problem, its the stuff i mentioned above. We will probably still need a few month in my team alone to close the loop.
And whlie we close the loop, models are continesly getting better.
ActionHank 59 minutes ago | parent
This isn't "AI is useless" or "AI is dead". They are saying it's not worth the cost.
Further, I would argue that coding is the only place it's made solid headway and even then it either scaffolds to less than perfect, or the code base grows to be unmaintainable. Not saying you can't ship a product with it, just saying that while you are moving fast and breaking stuff someone else is shipping a solid product marginally more slowly.
Last point "continesly" is not a word. Continuously maybe?
WillAdams 52 minutes ago | parent
I've tried to ask how much it would be necessary to pay to remove that, but haven't found an answer with a solid price.
I managed to get a local LLM running and processing such files --- it took 405 seconds to process two on my work-issued laptop (by hand, with the aid of a batch file cobbled together from a spreadsheet I can do ~200/hr.) --- I need to do almost 5,000 of these for Q4.....
freecodeio 41 minutes ago | parent
the_sleaze_ 30 minutes ago | parent
Or did you think he could have told the customer that MSFT messed up so no harm no foul?
opt3635 8 minutes ago | parent
that is exactly the value proposition of software as a service
goosejuice 42 minutes ago | parent
I find it fascinating that you could hold this position. Pick a single popular SaaS today and you can duplicate it in a week. Of course that's not the only hard part of running a business, you need to market it and sell it. But that's millions of dollars and months to years of work reduced to nothing.
Even marginal labor replacement in knowledge work leads to pretty significant revenue (not profit) figures split between the labs. I do think diffusion is relevant. It takes awhile to find product market fit and for new ways to work to become common place. We've seen this with most revolutionary technology (usually on order of decade+) and it's only been a year or so that these things have been good enough to replace some labor. I think we'll see incumbents who are slow to move on this get replaced by new entrants with much better margins.
Not to say the hype isn't fucking annoying and often bullshit.
freecodeio 27 minutes ago | parent
StableAlkyne 14 minutes ago | parent
The same cannot be said for SaaS: your business needs certain productivity tools in order to operate, and the only options tend to be these hyper-expensive SaaS options whose UX has been clunky since 2005. Historically, you could not build these in-house because it's more expensive to hire a full engineering team. Today, the math being done is whether it is worth it to build something 90% as good for 1% the price of the SaaS.
freecodeio 10 minutes ago | parent
clones are frozen in time and tend to rust and a financial incentive emerges to just purchase something that doesn't rust
mjr00 13 minutes ago | parent
This is a common claim but I have not seen it played out in practice.
The most well-known example people cite for this is Photocraft[0] which went viral a month ago; the claim was, at the time, that it was 80% feature complete upon release, and people took this to mean that Adobe was dead. The creator claimed that within a month there would be feature parity with Photoshop.
Currently, their feature parity tracker[1] has them at 43%; a whopping negative 37% improvement in compatibility. Yes, this was a change in measurement, but by their own admission:
> the 2026-10-03 estimate (82% "of the way") [...] counted feature surface and was far too optimistic
The impending "SaaSocalypse" was predicted earlier this year, but never materialized, and you can see it in the stocks of Atlassian, Salesforce and others. They're getting hammered in the past few weeks as with all other tech stocks, but if you put all your money into Atlassian in June you would still have tripled your money right now.
[0] https://github.com/storytold/photocraft/
[1] https://github.com/storytold/photocraft/blob/main/docs/targe...
goosejuice 10 minutes ago | parent
Of course. As I said it takes time. Writing the software is different than replacing an encumbant. A new software company starting with AI simply has less labor costs than any time in history. Claiming a competitor starting from the same place without AI is marginally slower is ridiculous.
ragequittah 39 minutes ago | parent
Does it cause problems? No doubt. People get way in over their heads with AI when they don't know what they're doing. But when you know your field and it's simply a conversation about busywork that needs to be done it's better and much cheaper than most assistants you would hire (unfortunately). It can save 10s of hours a week easily.
phoghed 25 minutes ago | parent
Why? The company you work for seriously doesn’t do anything productive with LLMs?
How about search? Every big company I’ve worked for has some shitty enterprise search. LLMs have massively improved that situation.
They’ve also almost completely replaced L1 support at my company, the numbers there are way down just from people being able to find information they need. I don’t have broad numbers higher level support, but for a couple niches that I know about they are down as well.
For example, in a global company of 100k+ employees you’d be surprised how many support requests HR got just for PTO related queries, not even real problems. Now we have an agent that can fully manage your time off, has the policy documents for every country, pulls your data from the HR system, their number of support tickets are significantly down.
Another project I worked on involved automating some risk analysis stuff as part of the IT process everyone has to do when they develop or procure anything. After developing it for a couple months, then a few months running human in the loop validating all of the assessments, we have analysts manually audit 5% of them now. All of this used to be a shitty excel process, now is a slop web app. Cycle time is down from weeks to minutes because nobody has to wait on reviewers and the app can prefill most of the data from your general system documentation. Overall everything is just as or more accurate,
I know these aren’t very glamorous, barely considered real work by some, but in my feeble imagination there are many such processes being automated and improved everywhere.
miltonlost 15 minutes ago | parent
Using Glean at work and it's not much better than bad searches. It frequently gives me outdated information, or uses questions I asked in slack as answers to the same questions I asked in the LLM.
metalliqaz 52 minutes ago | parent
This is not the only question.
As my own organization has rushed/struggled to implement AI workflows, I have seen this same sentiment in many places in slightly different forms. The inevitability of it. The obvious benefits. "The revolution is here." Ok. Except why is it always around the corner? Why does the investment keep growing but the returns keep pushing to the right? Why is the answer to all the problems just "more AI"? From what I can see, it is very much in question that these intelligence products can ever integrate (in their current form) into our organization in a way that pays dividends on the investment. I'm really not confident that the current products are going in a direction that will ultimately help us. Instead what I have seen is groups of users that believe they are much more productive, but their numbers don't actually reflect any improvement.
t-3 51 minutes ago | parent
lambdasquirrel 40 minutes ago | parent
petra 39 minutes ago | parent
reactordev 40 minutes ago | parent
…deploy.
(Sir, The airspace is shutting down)
Crap.
ratelimitsteve 38 minutes ago | parent
jrflo 55 minutes ago | parent
unglaublich 51 minutes ago | parent
pr337h4m 50 minutes ago | parent
inknight 49 minutes ago | parent
prometheus1992 48 minutes ago | parent
bilbo0s 37 minutes ago | parent
Baidu and Xiaomi don't create models. That's handled by other entities in the China market.
Baidu and Xiaomi experiment with models. Test the efficacy of different models. Get an idea of the relative economics of different models. Maybe create a model of their own to see if they can improve the business case in one dimension or another.
What they're saying is that, in their role as customers for these models, they're a bit confused about the business case. I think that's a fair statement. They've presumably put a lot into finding out how to make money out of these models, and presently, the returns are not there with respect to the bottom line. Additionally, their experiments have given them enough background to feel comfortable asking, openly, where exactly everyone else feels the returns will come from?
For them, yeah, just keep the little experiments going and don't go all in until someone offers you a model that can answer questions for less in cost than the person positing the question is willing to pay you. Maybe stay in verticals where the questions are valuable enough to be worth the cost for right now. Medical, Finance. And limit exposure to the larger market of less valuable questions: "ChatGPT, where is the closest pizza by the slice?"
mirrir 17 minutes ago | parent
12 days ago: https://huggingface.co/baidu/ERNIE-5.5-Flash-Base 20 days ago: https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL
locknitpicker 34 minutes ago | parent
If a leading AI company warns that there is a bubble, perhaps people should listen?
The likes of OpenAI and Anthropic have an interest in keeping the hype going, as losing the hype poses an existencial threat. They are highly unsustainable as-is, and losing momentum won't make that problem any better
> Also notable that Baidu/Xiaomi don't have any frontier models, so it's a bit like Samsung commenting on the AI race
Nonsense. The ugly truth is that nowadays the cutting edge frontier models might unblock niche applications, but the cheapest models in the Pareto front are already handling the bulk of the prompts/sessions.
And nowadays the only reason why cheap models such as GPT6 Luna are on the Pareto front is due to the fact that OpenAI rushed to slash it's costs in half right after it's release, in a desperate attempt to be close to the Pareto front and remain competitive.
I'm confident that once hardware costs return to normal, the bulk of heavy AI users will jump head first to local models and US companies will face an existencial crisis. There is a reason why US government officials accuse Chinese open weight models of being economic warfare.
ironborn123 17 minutes ago | parent
I am thinking maybe your blunt take is right. It is true their primary objective with distilling western models and open sourcing stuff has seemed to be to undermine the US economy. Cold war tactics, instead of running a normal business.
Otherwise per their own admission why would they invest huge amounts, that too on student models.
lenerdenator 53 minutes ago | parent
It makes total sense that they'd warn of this. They're also somewhat right. Not for the right reasons, mind, but they're right.
unglaublich 49 minutes ago | parent
999900000999 50 minutes ago | parent
Can’t send emails, fly or check your bank balance. Ohh well.
I would love to see 2027 be the return of QA across tech.
drstewart 44 minutes ago | parent
999900000999 34 minutes ago | parent
Seems connected.
slopinthebag 33 minutes ago | parent
duskdozer 10 minutes ago | parent
freecodeio 43 minutes ago | parent
slowin 44 minutes ago | parent
This has been revolutionary and game changing for me, but what it’s not is profitable in any way. In fact, quite the opposite. I also don’t even bother open sourcing or distributing what I build. It’s my computing dream, but it is just for me. Capitalism nor our idea of open source fit this model.
233mhz 37 minutes ago | parent
slowin 35 minutes ago | parent
I can’t foresee what happens to the economy, I just know that I’m far more empowered than I’ve ever been before, and I’m not going back.
233mhz 25 minutes ago | parent
There are maybe 100m people worldwide ready to pay $20 a month for a top of the line llm, no one will drop $2k+ on a machine to do the same at home with a gimped local model which brings 1% of the value since it won't be tightly integrated with anything
slowin 21 minutes ago | parent
Shitty-kitty 1 minute ago | parent
neogodless 4 minutes ago | parent
hvs 33 minutes ago | parent
meindnoch 32 minutes ago | parent
slopinthebag 31 minutes ago | parent
ahoka 22 minutes ago | parent
jedahan 25 minutes ago | parent
InsidiousWombat 20 minutes ago | parent
runarberg 9 minutes ago | parent
kamranjon 33 minutes ago | parent
dominotw 31 minutes ago | parent
what are these and why did you do this? I am the only one that doesnt need custom software in my life.
slowin 23 minutes ago | parent
It’s funny because they say the first 80% is easy, but the last 20% of polish to get a shippable product takes most of the time. Well, if you just build for yourself and never ship, you can just crank your own solutions out at hyper speed.
dominotw 3 minutes ago | parent
exitb 19 minutes ago | parent
SpicyLemonZest 5 minutes ago | parent
runarberg 5 minutes ago | parent
dominotw 2 minutes ago | parent
antiloper 18 minutes ago | parent
- A sane image editor for Linux, targetting the parts of Paint.NET that I need and skipping the rest (not having this annoyed me for decades)
- Closed caption overlay using a local nemotron-3.5-asr-streaming-0.6b
The latter I could've easily hacked together on a weekend myself, but I never had the energy to do it. Now it's (basically) free.
magicalhippo 49 seconds ago | parent
jstummbillig 29 minutes ago | parent
lopsotronic 18 minutes ago | parent
I'm not throwing shade on the entire risk field, but I AM throwing shade at betting trillions of dollars on what you think a tiny fraction of future costs might be going off into infinity. It's what made image sharing or bond insurance appear more profitable than asteroid mining or solar power, and it's the leprechaun promising the pot of gold under the AI rainbow[1] for financialists.
This happens to be a puzzle piece that fits pretty well together with several well-smashed shards of the Old European Right, but that's another subject and much more dubious. The "financialist thesis" of Western Decline is on much more solid ground.
[1] I.e., eliminating all future labor costs forever
12391827 11 minutes ago | parent
piva00 7 minutes ago | parent
This moral ship has sailed, as much as you and I would like for it to not be true, if this is the world we need to live in I will do as OP and use these tools to achieve stuff that wasn't possible with my free time before. In that sense I think it's an accomplishment to be able to generate code for ideas I had but didn't have the time to properly code them.
12391827 3 minutes ago | parent
SecretDreams 23 minutes ago | parent
I'm not seeing a lot of unique applications. I am seeing some optimization, plenty of rehashing the same crap. Lots of thought outsourcing.
Honestly, I'm mixed as someone who also uses these tools. Is it really efficient for 100s of people to be spending tokens and electricity to all make, effectively, the same bad app from their shower thought/idea? I love that we've democratized coding so that it's more accessible to laypeople - but it is mostly all slop and lack of creativity being produced. And, in the rare instances where someone genuinely makes something good, 100x people will then spin their own AI to rip off the same idea.
Where does this all go at the end of the day for humanity when nothing is new or creative and we are going backwards in critical thinking along the journey? Especially since the journey IS the thing for humanity - not the end product of the journey.
InsidiousWombat 20 minutes ago | parent
I expected tools when animator can just put a bunch of keyframes, sketch the transitions and have AI fill it in, getting all of the artistic vision with massive work saved.
Instead industry went full on dystopia "let's replace entire jobs" instead of settling on just massive efficiency improvements on top of keeping the quality and we got million slop generators and little of the actual good tooling that would change the industry, outside of coding tools.
surgical_fire 16 minutes ago | parent
That's the only thing the elites care about. They dream of not needing pesky labor anymore.
FLeXMurphy 10 minutes ago | parent
gilrain 12 minutes ago | parent
You could try, collectively, uplifting the voices who saw it coming and called it out clearly in advance. They're the ones generally derided and dismissed as impractical weirdos, though, so that might be hard.
As long as we instead uplift the greedy and craven, we will see greedy and craven results from otherwise neutral technologies.
cccbbbaaa 11 minutes ago | parent
dayvid 1 minute ago | parent
miltonlost 19 minutes ago | parent
220hertz 5 minutes ago | parent
uptrails 13 minutes ago | parent
On some level, I understand where the individual impulse is coming from - "I couldn't do this and now I can, it's amazing" - and I can understand why amoral VCs would throw money at these platforms. But I will never understand why so many intelligent, sensible people are cheering for this future.
Look at all the new book releases on Amazon. They are 95%+ slop, and not good slop. They're made with zero effort. Some of them sell because people judge books by the title & cover. What does this achieve? Once you eliminate all revenue for real authors, what next?
smallmancontrov 4 minutes ago | parent
Everyone learns the same basic curriculum in school. What does this achieve?
Everyone makes mediocre attempts to play instruments/sports as an adolescent. What does this achieve?
Everyone treads the same highs and lows of friendship and romance that have been tread by billions of others. What does this achieve?
It achieves the human experience. Not everything has to be best-in-class to be meaningful, and that's a dreadful expectation to heap upon anything that doesn't strictly demand it.
yipinwong 1 minute ago | parent