Phwoar.
AI and ChatGPT
- Thread starter JB08
- Start date
I dunno. Still looks like AI is used as a marketing catchphrase, while the heavy lifting was done by vision system? They do mention "deep learning" but only once, and not explained further, what I would expect should be explored more in the article. Feek free to disagree with me you Med-HC freaks.
Vision systems nowadays (to be fair for the last 10 years) are AI-based. In fact, regardless of the field, be it vision, language, audio etc, the architectures used are pretty much identical.
The vision stuff seems like the most obvious upgrade on humans, to my uneducated eye (no pun intended). Pattern recognition is hard (and incredibly boring) for humans but seems to be the way these things work?
it's amazing how everyone in the AI space has given up on actually focusing on ROI and financing. You look at anyone talking about AI IPOs and it's essentially people saying these companies are dumping their companies on investors so they can cash out
The "staff replacement" thing is quite funny to me because at my work I am 99,9% confident that no one can be 'replaced' by AI. They may become more productive with AI (AI can be useful) but there is so much nitty-gritty in our work that you cannot outsource that to a tool.
Canadian employers are paying the price after AI proves unable to replace laid off staff
Canadian employers are paying the price after AI proves unable to replace laid off staff
https://www.theglobeandmail.com/bus...e-paying-the-price-after-ai-proves-unable-to/
At the current rate of evolution of AI, even if not right now, maybe in a few more years. Theres not many roles which will be exempt.
China are even using AI powered robots to lay plaster and plasterboard now on large scale building projects.
Honestly we don't know if jobs get nuked to zero, if they get massive reduction or if we get mad productive and employment stays the same and we just produce more. They call the singularity for a reason after all....
At this point in time, I don't think jobs get wiped out
At this point in time, I don't think jobs get wiped out
What rate of evolution? LLMs and ML doing the same they've been doing for 2 years. Models have actually got worse in a lot of cases in that time period. More importantly providers are now trying to charge what they actually cost to create and run. It doesn't work because it's way too expensive...
Robots etc could be slightly different and there's been automation going on in manufacturing for decades. Fantastical claims about what robots can do or not needs to be treated with extreme scepticism, in terms of the work being done and also the whole cost of creating and maintaining such technology v just using dirt cheap Chinese workers.
Not really.
@anyone running agents
What models are you reliably using for default actions (some tool calls etc) and what models for escalations (heavy work)?
For me it goes = Deepseek V4 Flash > Claude Haiku > Deepseek V4 Pro > Claude Opus
Gemma4/Qwen3 for anything local/private.
What models are you reliably using for default actions (some tool calls etc) and what models for escalations (heavy work)?
For me it goes = Deepseek V4 Flash > Claude Haiku > Deepseek V4 Pro > Claude Opus
Gemma4/Qwen3 for anything local/private.
Last edited:
This is a "language" issue - or how you name things. You mixed "person being replaced" (your comment) and "AI to do parts of jobs" (linked article). But it boils down the same thing - more productive employees means "more (work) can be done with less (resources)". The quoted part was by the way EXACTLY how my company framed the restructuring (meaning firing people / moving jobs to other places), what caused a major outburst and they had to back off/change how they name things. The restructuring is still hapenning, no changes there.
If you have 10 employees in a same role, and you make them 20% more productive, you will either be able to let go of 20% of people, or you will be enabled to scale up 20% (in terms of volume of work they can process). This is hapenning now. Some companies are reducing staff, others are picking up more work with the same resources (like my org does).
Looking at this from individual perspective ("they can't replace me because I do XYZ, which can't be done by a machine") doesn't make much sense. It's more a question how much % of your work can be done by AI, and aggerating this to global will give us an answer to how big of a job crisis we're about to face in the next years.
Even roles like translators/graphic designers (both took MASSIVE hit due to technology development in recent years, AI or not) still exist. But it's almost impossible to get a job unless you're very specific/1 percentile of the best of the best. So, did AI "replace" translators/graphic designers or not? Like I said, it's a matter of how you frame it.
Personally I think waste in corporate jobs (copy-pasting between systems, or fixing issues generated because of bad copy-pasting between systems) is gigantic. Some of those are easy to fix with AI, technology is not a problem. People think AI is "not there yet" in terms of development to do a real impact, and I would disagree with this statement - even if we stopped developing AI in this second, current technology will do massive changes to how work is being done in the next years - because the limitation is adoption of the technology, not the technology itself. Of course, not every role, job, department or company is going to be massively impacted by it, but effect on the economy will be enough to make serious damage. And unfortunately, stopping development of AI is not on the cards, it's actually quite the oppisite - a race. And whichever company is the winner, it's the people who will be the losers.
Back to this quote:
This is actually quite an interesting conclusion. So what they did is
1) they fired significant amount of people across the org*
2) identified what isn't working anymore
3) hired people back to some roles (reshaping them at the same time)
*(and applied AI to do parts of their job - or not, we will never know if this is really the case)
They only had to rehire 1/3 of the people they let go, which is I guess not a bad number and a decent business case for other companies to follow.
There is a method we used in Software development in previous company: you create an app, you give it to users and once it's up and running, you start removing features/functionalities and see when do they start complaining (or usage goes down). That's how you cut down to the essential/core of your business. It's was a method when we didn't want to keep investing/developing the software, and wanted to reduce costs of hosting/maintenance. It's similar approach to the one above. Works if your goal is not expansion, but rather cost reduction. Shareholder is happy either way I guess.
Just made an entire iPhone app while having lunch. Complicated one too.
People saying this won't impact software dev jobs have their head in the sand at this point.
People saying this won't impact software dev jobs have their head in the sand at this point.
Was it approved by the App Store? Will you maintain the app by providing updates and bugfixes? How secure is it?
I'm sure it's a pile of shite.
Harsh. I’m sure the RateMyMickey app will be a great success.
I was guessing an app that predicts when he's about to get the scuts.
Is anyone able to get offers in the big companies / hot startups?
It is being really bad for me, the competition has gone extreme and I am completely unable to crack it. Able to quite easily reach the panel stage (only one failure there so far), but in panel, I am getting destroyed. Be it some random undergrad math question, some Leetcode problem when the recruiter specifically says it will be ML coding not Leetcode, or some AI question that is not in my domain. The only ones I am consistently able to pass are behavioral, in domain research, ML fundamentals and ML coding. DSA/Leetcode, Math and random ML that are neither fundamental or in my domain, and I get fecked. And regardless of the loop, it is inevitable that I am getting at least some questions like that.
Kinda depressing really. 3 years ago I was able to get half a dozen offers, this year, at this stage I am 0 for infinity and just another 2-3 loops remaining. At least when this is over, I will rest cause tired of multiple interviews every week for the last 2 months.
It is being really bad for me, the competition has gone extreme and I am completely unable to crack it. Able to quite easily reach the panel stage (only one failure there so far), but in panel, I am getting destroyed. Be it some random undergrad math question, some Leetcode problem when the recruiter specifically says it will be ML coding not Leetcode, or some AI question that is not in my domain. The only ones I am consistently able to pass are behavioral, in domain research, ML fundamentals and ML coding. DSA/Leetcode, Math and random ML that are neither fundamental or in my domain, and I get fecked. And regardless of the loop, it is inevitable that I am getting at least some questions like that.
Kinda depressing really. 3 years ago I was able to get half a dozen offers, this year, at this stage I am 0 for infinity and just another 2-3 loops remaining. At least when this is over, I will rest cause tired of multiple interviews every week for the last 2 months.
I know someone who is a newly minted PhD in ML from Toronto and they are saying the same thing - the job market for new grads at that level is brutal.
I think for fresh grads is even worse considering that there are millions of them. Really good ones I know are completely unable to even get interviews.
I am experienced hiring so mostly going for staff IC (L6) or Manager 1 level, so easy to get interviews but so hard in passing them that at this stage I am entering them without hope.
Some friend of mine who is a senior researcher (L5) just passed Mistral after 10 or so interviews but got heavily down leveled to the point of absurdity that he got angry and of course rejected them.
A position I interviewed for 6 weeks ago is still in LinkedIn. I literally crushed 5 out of 6 interviews but fecked the sixth and so they would rather keep the position open than hiring me (and I assume lots of other people have this experience).
Claude Fable released.
2026 pace has been staggering so far, Recursive Self Improvement (RSI) by 2028 latest?
Feels weird how lax Google has been but they have just signed a 11bn dollar deal a year with SpaceX for compute, perhaps they're going to try claw their way back.
Feels weird how lax Google has been but they have just signed a 11bn dollar deal a year with SpaceX for compute, perhaps they're going to try claw their way back.
Claude's tweet getting millions of views and my colleagues talking about it. Anticipation for the Mythos-class models is remarkable.
Shows Anthropic's PR stunt surrounding the dangers of Mythos was well thought out...
The visceral response is proof that people are burying their head in the sand.
I'll think of you all when rate my Mickey is top of the app store
The harsh genuine sceptics of ai ask evangelicals how ROI will be measured for companies using the products. They also ask how do Ai companies make a profit and the response is either "magic" or a non answer.
Who has their head buried deepest in the sand ?
Last edited:
I know that is not the point but still funny to see Fable answering ‘how many o’s in goggle’ especially if promoted to give a single number.
Run it multiple times in new instances for some added humourz
Run it multiple times in new instances for some added humourz
Yeah I'm impressed with Fable.
I guess i'm impressed too but i'm not doing anything complex. i'm seeing people do amazing things in my youtube feed though, like one shotting full games... so what happens now?
The movie itself is atrocious but it is seems to be the first full-length movie that is completely AI-generated.
I work for a big corporation who is going all in on AI, presumably because the CEO has drank the kool aid about efficiencies and is looking forward to making a load more people redundant.
I've recently endured a load of training on how Co-Pilot will make everything we do much easier and quicker. Talk about marginal fecking gains! Writing email, preparing for meetings, using Teams blah blah The difference between doing these tasks with and without AI help is so fecking minimal. All it really achieves is turbo charging corporate bullshit, screed of meaningless waffle spewed out before, during and after every meeting and emails twice as long as they would have been before, without giving any useful extra information. It's so easy to spot people using AI to do everything and I honestly believe they're not doing a better job or saving any time at all vs the luddites and sceptics. Although I can confirm that all of their outputs are a tedious ball ache to wade through.
Now I don't work for a tech company, and nobody I work with writes software, so definitely possible I'm not seeing corporate AI's main strengths but (based on what I am seeing) the idea that you can make loads of people redundant - then divert their salaries towards the tech firms pimping AI instead - and end up with a better run company is wildly misplaced. Not to mention that a pivot like that is absolutely terrible for society as a whole.
This makes me feel better. I'm going through the exact same thing. Now have an AI objective on my PDR. 10 hours of mandatory training on copilot.
Just saw this, curious about the pro-ai crowd explaining how the maths maths :
For me it, it definitely helps a bit, my SQL gets done super fast, I can get help with new features in programs I use and I can quickly get info from public sources. But it's like 10% gains.
I assume the logic is that 10% gains overall means you lay off 10% of your workforce and job's a good 'un. But that's obviously flawed, as you have to offset those savings with the (difficult to quantify?) money you pay the AI providers and you can't fully trust anything that AI comes up with, so you need those humans in the loop as a fail safe. The whole thing really feels like a house of cards.
Although I'm saying this as someone who has no fecking clue what SQL means. So what do I know?!
Can you trust the humans though? Dunno, definitely not in software engineering.
10% can mean layoff 10% of the staff, or it means you produce 10% of the output.
And I believe 10% gains in software engineering are a massive, massive understatement. Unless those software engineers were already the legendary 10x coder, the gains will definitely be higher than 10%.
What a bizarre statement in this context.
We both know that the software is full of bugs. We know that the reports done by humans very often contain errors (and very often the errors are deliberate).
It is the same argument of 'what happens when a self-driving car kills some human' while ignoring that humans driving cars kill more than a million humans each year.
I understand the arguments against AI (and if it was to me, I would stop AI research in almost every field right now), but 'can we trust AI' when humans are full of shit at whatever they are doing is a non-argument IMO.
My current take is that large corporations are going to really struggle to integrate AI into their existing workflows and will miss out on the big productivity improvements. There's just too many bottlenecks, legacy systems, regulation, and stubborn humans in the mix to change things quickly.
What I do expect is that AI is going to make smaller firms and start-ups, with tiny teams and budgets, much more able to compete with the big incumbents.