Posts by frankyunderwood123 342 publicly visible posts • joined 25 Jun 2019
you give a prompt to an LLM under the same circumstances, you get the SAME output
That’s not strictly true unless you really go some with prompting.
LLM’s are non-deterministic and with the frontier models we’re dealing with trillions of parameters.
Sure, you’ll get results which can be contextually the same.
Sadly not, because Claude.
However that was freely admitted so nothing being hidden here.
These days you really have to go some to impress with Doom, there has to be some real world connection.
Doom powered by 100 mouldy potatoes, or running on gut bacteria.
Both real projects. Both incredibly impressive.
At the huge mega corp I work for, one of those near invisible ones hardly anyone has heard of, we’re currently in experimental mode, but a fast tracked one.
A lot of devs here are super aware this free lunch is going to lurch abruptly to a halt.
My take is this is when the redundancies will occur.
Rather than blame silly token over usage on deterministic tasks, there’ll be a headcount reduction.
Can’t have the coolaid drinking C-suite having egg on their face over rapid adoption, can we?
Anyone not onboard due to “using it wrong” will be ejected.
A few years later? They’ll be crying out for tech talent to dig them out of a shit storm.
Closed loop water cooling should be mandatory.
Large Buffer zone between site and residential areas.
Light pollution level requirements should be stricter - no reason for stupidly bright illumination around the facility.
No slop farms and offer incentives for data centres that benefit humanity, e.g. cancer research and other health research, alternative energy research etc.
Then again, I don’t expect anything even remotely as sensible as this to ever happen, especially given AI tech leaders are currently haemorrhaging money and the cost of energy in the UK is not competitive
… to engage with linkedIn for anything other than a basic profile and a CV.
Sadly the site is still almost a requirement when job hunting, but any trust of the platform beyond reviewing a CV would be foolish.
It’s not just AI slop articles though, it’s AI slop profiles and CVs (resume for US folk)
The last time I used the site for my own profile was 8 years ago when job hunting, which was a waste of time. Got a job via word of mouth as per usual.
Even back then it was mostly rubbish, a site full of banality and back slapping, duff articles and fakeness.
I was recently helping interview candidates for software engineering roles and found myself back on the site.
It was so depressing. Everything seemed AI generated, down to the candidates profiles.
Quite why LinkedIn still has the draw in terms of having to use it when job hunting is beyond me.
I really felt like giving negative points to any candidate who put a linkedIn URL in their CV, sadly every single one did.
As a developer being forced to use agentic AI and SDD, it’s clear to me now that we are facing a disastrous future of software instability - we are teetering on the brink.
The rush toward agentic AI despite the risks is… very human. Greed and power.
Anyone noticed the instability yet? Beyond the fairly usual we’ve had since computing arose?
It feels like there’s this huge unwieldy dam wall , a virtual one, about to collapse.
Buckle up, this will not be fun, but perhaps there’ll be some high paying jobs to clean up the sloppy shit after the bubble bursts.
If you are unfortunately forced to use windows or have to use it for a specific purpose then the suggestion is you install some anti-malware software for windows, otherwise known as a debloater.
There's a few well eyeballed open source repos that you can find if you search about.
Malware:
"Malware, short for "malicious software," is any program or code specifically designed to disrupt, damage, or gain unauthorized access to a computer system."
Windows falls under Spyware and Adware, which makes it Malware.
Another suggested fix is to use a different operating system that may be able to run the software you need for that specific purpose.
It could be an obscure app or even a popular one that only runs under windows.
It could be a game.
This fix is gaining traction - rapidly approaching well into the 90% of windows apps/games running flawlessly on other operating systems.
(e.g. Linux or macOs)
Sadly if your company issues you with a windows based computer there is no solution to the malware problem - just be thankful you have a job OR ...
Try and trick your friendly sysadmin into allowing you to install virtualization software...
If under Burnham the bit barn AI “we can’t afford to miss the bus” continues, that will be severely dissapointing.
It’s a race to nowhere but lost billions for the UK and a slap in the face for any plans to improve peoples lives.
The UK should sit out the AI bit barn building until after the coming AI crash and then assess the best course of action.
You can’t build hundreds of AI bit barns AND hundreds of thousands of council houses and magically expect the grid to cope, not to mention the water supply!
Really? Then explain the 2016 US elections.
A better way to phrase this could be adults had more access to learn to think critically as they were growing up.
Clearly, given the state of the world, not many adults bothered.
I’d say the ability to truly think critically is a rarity.
… who suffer from frequent crashes of teams or outlook or whatever are on win11.
Those on macOS and Linux have no problems at all.
It wasn’t like this with win10.
This is anecdotal, sure.
They are all on reasonably specced Dell laptops, which actually work amazingly well with Fedora installed.
It seems to me that Microsoft continue to hold dominance for their OS in the business world simply due to their other products. I’m sure there’s all manner of deals with big corporates where the OS for all intents and purposes is pretty much free - bundled with office licenses.
Back when I used windows as a primary OS for personal use , the only copy of windows I ever purchased was the one that made me leave windows for all but gaming, the ill fated windows 8. I still have the box and disk, perfectly possible to use the key to get win11 pro installed with just a quick phone call.
The OS has been free for decades via one way or another.
It is even possible to get a reasonably sane install of win11 with easy to use debloat tools, but you remain at constant risk of microsoft pulling the rug from under you with updates. Pretty much like running a hackintosh, where an apple update could ruin your day.
I no longer have a windows partition for games I cant run as I stopped VR gaming a year back.
Feels so good.
The only “benefit” from this insanity is the big tech may possibly profit in the long term. For everyone else, which would be 99.99999% of the planet, it’s nothing but a loss.
When the AI bubble bursts all of us will be picking up the tab.
In the interim more jobs will be lost than gained.
I’d love to know how much of the energy being used is for genuinely useful tasks and how much is for pointless slop generation.
I’d happily wager over 95% of the power consumption is wasted.
Yet this madness continues.
I find myself hoping that the UK struggles and fails in this game, that red tape, high energy costs and dithering sets us back, followed by the inevitable crash.
If the government under Burnham pursues an agenda to try to compete, they will be scuppered.
We all know that a right leaning loony party like Reform, if in power, would put it at the top of the agenda, selling out to big US tech and promoting fossil fuels.
What a waste, a failure.
We don’t need it in the way it’s currently being marketed and sold.
But as a valuable tool you’d have to be a complete idiot to discount the value it can give.
The problem is that it’s being applied to everything without any deep analysis as to whether it is needed.
Token cost will surely galvanise minds, as will a severe market correction.
Right now businesses are wetting their pants over the prospect of employees being able to orchestrate agentic AI to triple, quadruple or even 10x their output.
That’s hokum. Not. Going. To. Happen.
There’s this huge what I like to call janitorial overhead, where you have to go back in and check the output.
How many of us are going to want to start multiple agents going and be the janitor for multiple threads?
How many of us really have the mental capacity to do this whilst really checking up on output?
“I’ll just leave this running and nip to the shops quick. “ is the more likely reality.
For complex tasks a huge part of those tasks is very often breaking it down into smaller chunks. Until one part is done, it can be difficult to start on the next.
Also more often than not there’s a planning bottleneck - there ARE no more tasks to work on until a milestone is reached because so many companies are crap at planning.
It requires context and AI is crap at that still. It gets very expensive to carry all that contextual baggage around.
I’m always shocked by the broadband divide in the worlds richest and most powerful country, that also hosts the worlds biggest tech companies.
Sure, it’s a huge landmass, equivalent to the EU and yes, overall it has better broadband access … but, it’s the home of big tech?
Asia dominates in fast internet availability per capita.
Why doesn’t the US?
… but sure, if LLM’s actually conceal what they have in their training data and add in a few porkies then they’ll be even more human sounding.
Most “facts” in this era of anxiety are lies or based on them. Post truth world is totally a cliche, bit like many, there’s more than a grain of truth.
The problem would really come in when LLMs are politically motivated and start spewing out FUD to suit a current regime.
… and that IS going to happen and likely already IS happening with some models.
the trend for lazy and verbose code due to more powerful compute has a long history.
Equally the trend for continuous “improvement” in software with more and more features added is an old one. You have to admire software that is “done” and the only updates are bug fixes and upgrades to suit OS updates.
Perhaps the increasing cost of hardware, of compute, may prompt some sanity back into the art, or perhaps not. Feels like the genie is out the bottle, but kudos to Dave for showing the way!
I may no longer use windows, but I’m subscribed to Dave’s channel. The man is a legend, a wizard, a sage!
The main subject is about Java, sure, but the underlying reason is that Apple are dropping support for intel, hence the decision by Oracle and others.
If Apple decided to support intel macs for longer then it stands to reason Oracle would provide Intel support for longer.
... although that is really close.
The last Intel mac released was 2021, so that's about 6 years of support.
Not brilliant, but par for the course when it comes to Apple.
This will also mark the end of the Hackintosh or close to the end.
I'm sure some will continue on for a few years after, but with no core OS security updates, it'll start to get risky!
I've been a macOS user for close to 15 years now and started on a home built Hackintosh - such an amazing project.
It'll be really sad to see it die, but there isn't a snowballs chance in hell of it continuing in a secure way.
there’s little doubt a crash is on the near horizon, the question is how to mitigate the financial fallout as best you can.
timing the market is a fools game.
I’m very close to retirement, so I’m considering an annuity or moving my pension into lower performing but more secure assets.
This crash is set to be the most punishing ever, given the insane amount of capital chasing a single technology. It reaches into every single corner of the economy.
for personal use, at least, windows 11 is avoidable, relatively easily, assuming you have a basic grasp of installing operating systems.
This may be in the realm of the computer geek, but outside of that realm, personal computing is dominated by android and iOS making it a moot point.
we’re in such a different world to when windows 7 was released. Back then most homes had at least one “beige box” and in wealthier homes a few more.
I reckon in the vast majority of homes for the last decade, those boxes have either been sent to the tip or are gathering dust in the loft rather than being replaced.
The primary market is business use and if you are unlucky enough to be forced into using windows, I’d argue it doesn’t make one bit of difference over win10 or even win7 - your work computer is going to be locked down and you’ll face exactly the same frustrations as you always did.
Does anyone really believe that Linux would fare better in such an environment amongst users who are not computer enthusiasts?
That Linux would be somehow immune from issues?
If you do, I have a bridge to sell you.
Fact is win 11 is just about good enough and that’s always been the case with microsoft.
because Amazon clearly hates those pesky employees stinking up the place.
It was mostly just the warehouse workers who demanded too much, like bathroom breaks and competitive salaries. It’ll take a while longer to get rid of them, but our robotics team is making great headway now 90% of them have been replaced by AI.
In the meantime, however, there’s all these annoying entitled software engineers, the ones we didn’t replace with cheap asian devs who never complain.
They have to go. It’s now super easy to get rid of them.
Soon it’ll just be the C-suite left, apart from the CFO and maybe even the COO.
… and yet this nonsense idea is being taken seriously?
we can’t stop a crime wave of theft in supermarkets.
we can’t control our borders properly.
we can’t do government IT projects on time and budget and effectively.
the list goes on…
yet people expect this government to convince big tech, big american tech, to facilitate device scanning on all mobile phones?
it isn’t going to happen so just calm down
whilst there’s a skill in prompting effectively to get better output, after a year of working with LLMs I concur with the sentiment of this article.
If you work at more or less the same pace as you did before AI, you can get better output, by making AI do the grunt work whilst you focus on the hard stuff.
But the business doesn’t want that, they’ve been promised massive productivity gains and better bloody we’ll see it given the increasing cost of tokens.
So corners are cut, skills are written that generate extremely verbose, repetitive code which makes it into production.
Given much of the increased output of LLMs has only just reached a point where it feels acceptable and software projects can take years to launch, we still don’t know what the outcome is going to be ..,
But we can certainly hazard a good guess.
LLMs can produce reasonable quality code when promoted effectively, but they can invariably produce code which on quick inspection looks good but is flawed. Because the code is often so verbose it is painful to review.
There’s also the temptation for engineers to churn out massive pull requests instead of small units of code, so you end up with review fatigue.
The slop is going to hit the fan in the next few years, there’s already signs of it from companies who have had access to the most recent models for longer, such as microsoft.
I fear this is just the tip of what is going to be an iceberg of failing code.
I’m retiring in a few years, so for me it’s sit back with the popcorn and watch it all unfold…
… but with some concern about my finances. Will the slop code reach the financial sector and cause a massive fubar that wipes out my pension?
“The phone, today, is at the center of your digital life and therefore everything is around the phone,” he said.
Speak for yourself mate!
"And of course, Amon expects, 6G networks will supercharge these sensing capabilities. "
Oh, that's ok then, it'll be 20 years before 6G has 99% coverage (that's really about 50%) in the UK.
“If you have smart glasses, they see what you see, so the connectivity needs to enable a very fast uplink,” he said. “6G is going to make all of us into walking cameras in this world.”
Yep, if you want to be a glasshole, you just go ahead. The rest of us will shun you.
This guy is just doing schilling of Qualcomm - what he is proposing isn't even remotely feasible in the real world.
It may work in a little Silicon Valley bubble, but at scale? - no chance.
The cost, complexity and social upheaval are so immense, it'll take decades for something like this horror to happen and let's all pray it doesn't.
If it even gets close, the kick back not just from individuals, but businesses, will be sure to scupper these insane plans.
"Sorry, we can't afford any more <insert-service-here> licenses - budgets are tight!"
"We're going to migrate everything off <insert-service-here> to cut costs this year - budgets are tight!"
"You'll just have to hold onto that old laptop of yours for another few years, there's no IT hardware budget this year!"
....
"Hey, we want you to extensively use AI to do all of your work via agents. Here, have a Copilot license and Claude Code one too."
"We expect you will be 10x more productive!"
"Use AI to review the code - but with human oversight, of course."
....
"Er, excuse me <team-mate>, your PR has 123 new files, 55 deleted files and 323 changed files in one commit - can you please break that down for me? Ah, whatever, I'll get AI to review it"
....
"We've noticed our bills are sky high and the quality of software has slumped, we're going to have to lay some of you developers off. Sorry, we'll try and reassign whoever we can whilst we restructure your department."
so are you saying non-AI code is infallible?
Clearly it isn’t.
I’m no AI fanboi, but used as a TOOL with considered prompting it can and does automate grunt work.
With the right governance and skill it’s also possible to orchestrate multiple LLMs via agents effectively to automate more complex tasks.
Yes, there’s a frikking boatload of hype around AI, but that doesn’t mean it’s all shite and it also doesn’t mean it isn’t going to improve incredibly quickly - it already has.
… is not paying attention.
You may well be a skilled engineer or equivalent that can integrate multiple agents to increase productivity and are perhaps what these types of companies want now.
How long do you really think that will last?
The pace of change is so astonishingly fast , if you are the one with ace AI skills you are very much part of training the next wave of agents.
Anyone using big tech AI services is giving valuable data.
AI slop? yes, the internet is awash with it and its easy to generate slop with AI tools, but if you know how to use them, that isnt the case, no matter how many el-reg greybeards downvote sentiment like this.
Agents in the right hands are bloody scary sometimes in terms of how much they can do to save time.
They will get better.
There will be catastrophic IT disasters ahead as a result of AI use, there’s already noticeable degradation in, for example, microsoft products, but I will bet it will ease over time.
AI bubble going pop?
Yeah, that’s baked in, isn’t it?
But out of the dotcom bubble burst we ended up with these scary titans of tech, these tech overlords if you will.
Something very close to AGI is just a few years away. That would be enough to pop the bubble and totally change the game.
That means even AI literate whizkids are at risk of redundancy real soon.
Think your decades of knowhow and experience can save you?
For now, sure.
But not for long.
You are the equivalent of the skilled artisan just as the industrial revolution started.
/checks 4G connection speed - 33mbps - good enough as an emergency backup at home.
Given the amount of wifi coverage across the UK, I'm struggling to think of any reason to use 5G for my personal use cases.
4G is coverage is still appalling in so many places, so why would 5G improve on this?
The answer is 5G wouldn't improve on it.
We've had 4G since 2012 - that's 14 years for the networks to get their shit together.
That's supposed to provide 99% coverage - I'd love to know the reality of that.
I think the reality is outdoors, in a field next to a mast, holding your phone up in the air on a sunny day with no wind = 99%.
Sitting inside somewhere? 50%?
Most of us still rely on WiFi in public places due to the spotty nature of coverage.
5G coverage is supposed to be in the region of 80% - yeah, right, good luck with that one!
And it'll be even WORSE indoors.
6G is a pipe-dream for the UK.
… to read about all of this because I’ve yet to experience it and have thus far avoided even seeing any video footage or screenshots.
I’m fortunate to use macOS at work and while I have an office suite installed, I avoid all but email and teams video calls.
However, having said this, all is not rosey at the big corp I work for.
AI is being rammed down developers throats in many many other ways.
No escaping it, so I just embrace it.
You want me to use AI for everything?
Okeydokey, I’ll just sit back in my chair and ask claude, by way of my microphone to do my job for me.
In the meantime I’m off for my mid morning nap, followed by a bit of light gaming, a walk to the shops and some TV.
retirement is great ! /s
Workday, for those that have the joy in using it … /s, is pants for the end user - the employee.
Links that open modal windows with the same link in it, secondary navigation attempting to be contextual yet lacking a way back. Multiple ways to discover the same section - and discover it is, as the navigation is absolutely shocking.
But who knows, perhaps AI could improve things, just so long as the ENTIRE UX/UI management team is shown the door, if they are given the LLM keys , expect an even worse user experience.
I reckon it’s just a matter of time before the same happens at the big corp I work for.
AI adoption is in overdrive, rammed down employees throats, you WILL use this. You WILL build new AI tooling.
Right now the motive is all about freeing up engineers to do more important and “rewarding” work.
The reality is engineers are building the tools that will make many of them redundant which will inevitably lead to just senior architects overseeing all engineering.
It will be fun seeing some pointless managers being the first out the door though.
don’t you just love the fact using browser text size adjustment tools has no impact on the text size on the home page?
That is such a massive fail.
It works on article pages though.
I struggle to use the site on my phone, my eyes aren’t what they once were.
They just need to do a few hours work so that there’s only one column below 640 and bump the text size by twenty percent.
problem fixed
In hands of a senior software engineer who has an understanding of the strengths and weaknesses of the tech, AI models are now generating decent code.
That engineer also needs to understand the code produced.
The process still needs to be managed with planned code iteration and tasks broken down to ensure manageable pull requests.
Going further, a deep understanding of how the architecture works and integrates with other systems is essential.
Does this always happen with AI coding?
NOPE.
What many of are seeing are pull requests with hundreds of code changes and worse still, brand new code with hundreds of files.
Not even the basics have been attended to, such as agent instructions. You end up with massive rambling functions with multiple parameters and no attempt at DRY or even the basics like code linting.
When asked to explain what the code does, the “coder” will regurgitate verbose AI generated explanations and you just know they are clueless.
Without good governance, this slop code makes it into production.
It works until it doesn’t, usually because it’s unmaintainable, not type safe and a change is made elsewhere which that code cannot handle.
And often nobody knows why.
This is not unusual even outside of AI coding, it’s just that now we can expect a tsunami of it. An unmaintainable wave of slop impossible to review.
“yeah, looks good to me” says the senior dev in their 30th PR review of the day…
The bubble will pop in terms of hype and a lot of pixie dust vendors will fold, but the underlying tech is going to remain.
There will be winners.
I’m continually amused that many grey beard devs lump all of AI under the same lazy trope of slop generators, almost as if they’ve never actually used the tech.
Yes, there’s tons of hype and tons of slop being churned out, but some of the tools are now so capable it’s downright scary.
AI is with us to start stay I’m afraid. I wish it wasn’t so.
I see no excitement in AI.
But I'm also not a grey beard kicking it with a "AI slop! AI slop!"
It can be incredibly adept at coding in small increments and can be a total genius copilot.
Anyone who has actually used it in a real world setup where a corporate entity is footing the bill knows that.
But it's also damn depressing and demoralising.
That cool little function you spent hours or even days honing into a beacon of brilliance, getting much kudos from your peers?
AI will have slurped similar up - stolen it - and will spit something very similar back out with a single prompt.
Will it be as elegant?
Maybe, maybe not - certainly over time the models have improved.
But what's the point in coding it yourself?
What's the point in spending those enjoyable hours iterating on your initial ideas?
Some junior dev is going to vibe code something similar in 60 seconds.
Worse still, a senior dev who knows exactly how to prompt will generate a far better function in 5 minutes.
We're being pushed into using AI because if we don't, we WILL lose our jobs.
But we'll eventually lose our jobs anyway.
The interim zone is what I'm currently sitting in.
Being asked to review a 192 file pull request copilot generated in "TypeScript" with 126 instances of using "any" as a type.
I'm flat out refusing to review that slop.
Slop because using AI to generate that amount of code right now is just dumb.
Small chunks of code in a PR - yep, AI is just a tool for good coders.
For vibe coders? It's a time bomb.
And that's why it's so damn demoralising.
It has been a wild ride, hasn't it?
To think that 15 years back as a software engineer you had a golden ticket - you could pretty much command an incredible salary as a senior developer.
Way better than a salary in middle management, sometimes close to senior management and way more fun.
Sadly, all but over.
I'm clinging on until my pension can afford me a slightly higher quality of life and ploughing nearly everything into that pension, but if I last another 2 years, I'll be lucky.
I'll either throw in the towel out of frustration or hopefully be made redundant and get a fat pay check to bolster my pension.
This is very London centric.
I think London should be considered as a different country to the rest of the United Kingdom.
In most places outside London and other large UK cities, there is no effective Public Transport.
That means we don't even get the luxury of the potential of a train arriving.
If you are "lucky" enough to have a train station within reasonable distance of where you live, the chances are it'll be miles from the nearest pub and/or anywhere really.
You then get the pleasure of paying the highest ticket fees in Europe, if not the entire universe, for a service that isn't fit for purpose.
My nearest railway station - and I live in a large town - is inexplicably 6 miles away near a tiny little village.
To walk from that tiny little village to the train station is about 15 minutes and half of the route has no pavements in a 60mph section of road with no street lighting.
Also inexplicably, the last passenger train left the now closed and indeed demolished and built over railway station in my town 60 years ago.
To walk or cycle from my town to that "nearest" railway station is about 2 hours along country roads with no pedestrian infrastructure = risk of death.
Trains at that station cost 10 times more to travel to my nearest city than the local bus costs.
Also, even more inexplicably, it's always raining when I try and walk to a train station - I have no idea why.
Yay Britain! Way To Go!
We're in for a wild ride and God only knows what mess will be made before this one sorts itself out.
Agentic AI is a recipe for disaster without safeguards, but since OpenClaw was unleashed, it seems safeguards be damned in big tech bro world.
What's alarming is how a single dev over a weekend pretty much beat billion dollar companies when it came to unleashing the obvious.
The lack of control is alarming.
Deeply alarming.
That OpenClaw allows a random Stan with a few thousand dollars of tokens to burn to unleash agents and attempt wreaking havoc across the web?
Sure, one random Stan's agent should be effectively blocked.
100 will. 1000 will.
But multiply that by 10,000 Stans with agents?
One hundred thousand?
A million?
We're about to enter an arms race and compute requirements are going to go off the frikkin' charts.
Websites are going to struggle to cope with a tsunami of AI agents scraping up content, blindly and stupidly throwing huge globs of data around, sucking up insane amounts of power - a slop storm of epic proportions.
Buckle up.
Celebrate, software engineers!
We'll have work for decades to come, fixing technical debt, because it's such a fulfilling use of time - who wants nice new greenfield projects? /s
They say that each revolution threatening to put people out of work eventually ends up generating as many jobs.
Now we know that AI is no exception.
AI can detect the technical debt and us humans can prompt it to fix the technical debt.
We get to spend our time reviewing AI generated code, rather than coding ...
... because, you know, we didn't get into this software game because we enjoyed the art of coding, did we? /s
