Will LLMs take your job? Only if you let them
- Reference: 1684139409
- News link: https://www.theregister.co.uk/2023/05/15/will_llms_take_your_job/
- Source link:
It's no good looking elsewhere for help. In the UK, the Minister for Education seems to be the spokesperson for official AI appreciation. She says that LLMs can [1]ease the burden of teachers by doing drudge tasks like marking , but if you think that sounds monumentally stupid she confirms it by admitting that she's got no real idea what's going on. A new technology is threatening to restructure employment and the government is barely bothering to even busk it.
We've been here before, of course. In the mid to late 1970s, the microprocessor stopped being a quirky outlier of a computing industry that was largely isolated from everyday experiences. Instead, people realized that microprocessors were going to infiltrate everywhere and change lives across the board – especially jobs.
[2]
There wasn't much sign of this realization in the UK governments of the time. They thought in silos: technology was about technology companies, and Britain's tech industry was a mess. Consumer electronics had been ceded to imports, the defense sector sucked up Cold War finance and made little of interest for anyone who didn't have to buy it. The same was true of wannabe-IBM ICL. Managed decline and the exit of the state from industrial influence was what counted as a tech policy. That a revolution was coming that would take tech into everyone's lives, potentially automating away millions of jobs, was no more a state problem than writing limericks in Sanskrit.
[3]
[4]
Then something remarkable happened. The BBC made a documentary, 1978's [5]Now The Chips Are Down , as part of its well-established Horizon series. It covered what microprocessors were doing and could do across society and industry, and unusually for Horizon – and the BBC – said the UK was adrift, there was no political strategy, and there had to be one. The ending was unusually forceful: "What is shocking is the government has been totally unaware… the silence is terrifying. It's time to talk about the future."
This hit home. Three years later, the UK government kicked of the [6]Microelectronics Education Programme (MEP) and backed the BBC's Computer Literacy Project – which saw the adoption of the BBC Micro as practically the standard school computer. None of this cost very much, the MEP was funded to roughly £35 million in today's money, but it helped fuel a national obsession with learning about, playing with, and programming computers.
[7]The first real robot war is coming: Machine versus lawyer
[8]The Shakespearian question of our age: To cloud or not to cloud
[9]The UK's bad encryption law can't withstand global contempt
[10]ChatGPT has mastered the confidence trick, and that's a terrible look for AI
Lots of expected things didn't happen, such as a viable educational software market, but things like Arm and the UK video games industry made up for that. How many people were put out of work by automation is impossible to say, compared to those put out of work by mismanagement and industrial decline. The service sector grew at a rate unimaginable without the use of IT, and economists will grow fat for years to come arguing exactly what happened. That the country was primed for action, though, is indisputable.
What would the equivalent for LLM knowledge engineering look like? This time, the jobs at threat are thought to be lawyers, finance advisers, coders and the like, especially those lower down the food chain. As with processors, the conviction that change is coming is as strong as the admission that nobody knows what it will look like.
[11]
A crucial difference is that what processors do well and do badly is clear. LLMs far less so. They deal in knowledge but have no idea about truth. They mix facts and hallucinations. They amplify bias. They are great at talking to people, but have no idea what effect they will have. In short, the technology is far less mature than its breakneck adoption would imply, and how things will evolve over the next five years is unknowable. This, frankly, is our salvation.
The one aspect of LLMs that can't be denied is that they are far too flaky without a human in the loop. A whole new field has appeared overnight, prompt engineering, based on the massive correlation between the usefulness of the output and the cleverness of the request. People who are good at their jobs are those who know the right questions to ask and how to interpret the answers, and LLMs can't do either. They can only be a tool of greater or lesser help to the human.
So here's the plan to make ready for the "AI Revolution." Expose students to the technology, not to replace teachers or encourage dependency, but to be skilled in critical analysis of what it does and how it does it. Encourage those in work that may be affected to find what each new iteration is good at and bad at for their particular specialties. This has three facets: yes, LLMs may be good for you and you should know how. No, LLMs can't work without expert human interaction, and for your job you are the best expert human. Finally, you need to know how to tell your boss that they're full of it if they try to boot you out in favor of ChatGPT.
[12]
It's unlikely that the governments and public service broadcasters of 2023 will find common cause this time to create even the skeleton of an education and awareness campaign. That's OK. This time round, the microprocessor has given us access to all the things we need to do it for ourselves. The scale of the state's involvement at the time was modest; it doesn't take much to have an outsized effect.
If you're smart enough to do your job better with LLMs or smart enough to know why that's not happening, you're smart enough to keep it. Don't wait to be told. Not that you would – you're a smart cookie, after all. ®
Get our [13]Tech Resources
[1] https://www.thisislocallondon.co.uk/news/national/23508293.ai-take-much-heavy-lifting-teaching-says-education-secretary/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2ZGICwTFO@RXkP3AoXJr8egAAAAU&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZGICwTFO@RXkP3AoXJr8egAAAAU&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZGICwTFO@RXkP3AoXJr8egAAAAU&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://www.bbc.co.uk/iplayer/episode/p01z4rrj/horizon-19771978-now-the-chips-are-down
[6] https://en.wikipedia.org/wiki/Microelectronics_Education_Programme
[7] https://www.theregister.com/2023/05/09/the_first_real_robot_war/
[8] https://www.theregister.com/2023/03/20/column/
[9] https://www.theregister.com/2023/03/13/column/
[10] https://www.theregister.com/2022/12/12/chatgpt_has_mastered_the_confidence/
[11] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZGICwTFO@RXkP3AoXJr8egAAAAU&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZGICwTFO@RXkP3AoXJr8egAAAAU&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[13] https://whitepapers.theregister.com/
Overly optimistic
...prompt engineering...
That is a euphemism for prompt slave .
They [LLMs] deal in knowledge but have no idea about truth.
The LLMs have no concept of truth. LLMs have no concept of anything whatsoever. LLMs cannot conceptualize and therefore can not reason.
Replacing people with LLMs is like telling bacteria to come up with the theory of gravity and space-time. You can do so, but it will fail to take you to the sought after results.
"If you're smart enough to do your job better with LLMs or smart enough to know why that's not happening, you're smart enough to keep it."
Yeah but managers are dumb enough to just arbitrarily pick the staff they want to keep and which ones to bin, when AI starts shrinking workloads, it's the weirdos that get the bullet first, which in an industry full of indispensable weirdos is a bad thing. Managers and higher ups don't sack people that let them win at golf. Know what I mean?
Doing a good job and making a higher ranking member of staff look good are not usually the same thing.
As techies, we need to be finding ways to use AI and LLMs to replace these bellends before they try and replace us. Project management is a big juicy target at this stage. There is absolutely no reason we can't use AI to automate that. Same thing applies to CTOs. Why not make the "technical" part literally true...it can't be worse than the sheer number of human CTOs out there that have never deployed a single box in their life / written a line of code since 1987.
I swear, if I spent less time "bringing them up to speed", I'd probably get a productivity jump in line with using an AI to assist me.
As a person that has worked as a CTO, project manager and general management shadow on and off for many years (which is essentially the same sort of thing as prompting a beta version of an AI trained on a dataset built in the 80s), I'd dearly love it if I could be the shadow for an AI. I'd be doing the same job, but I'd get much better results.
Quite a few of you out there will be working as a shadow and not know it yet...some pointers to help you figure it out.
1) There is an important executive meeting coming to discuss some sort of technical strategy. You aren't invited, but each of the attendees comes to have a chat with you on the sly individually to ask your opinion on things that have very little to do with any of the projects you're working on.
2) You hear the phrase "off the record" a lot.
3) A weirdly high number of suggestions that you make "off the record" seem to get implemented, but never accredited to you.
4) You're low ranking, but basically running the business.
5) You know exactly what was discussed in a meeting without having been there and you start to control these meetings by providing one person with one set of opinions, mostly ones that you placed there to make the person look smart, but a couple that you agree with, and you arm another person in the meeting with counter arguments to cancel out of the fluff you gave the other person, thus controlling the outcome...without setting foot in the room.
6) If these people are being too cagey with information and you can't quite execute a full plan, and you need more details...you engineer the information that you give these people (basically control the meeting to come to know particular conclusion) to lead to them requesting you join the meeting half way through at which point, you gather a little more information and tip the balance where you see fit.
7) People start sweating when you try and book a 2 week holiday.
I swear, I sometimes feel like the BlackAdder of tech...I've always got a young Baldrick that I'm training up and I'm always deploying cunning plans...it's never been limited to just one company either...I've managed this at several businesses. Mostly because the Baldricks tend to move on and eventually become managers, which opens the doors elsewhere for more cunning plans. They don't always work, but they are usually, so cunning you could brush your teeth with them.
As for AI and how it would help in these situations, I'm not sure it would per se...but past a certain level in most tech firms, you'll find that it's already automated to a degree by someone like me.
Remember kids, puppets are controlled by strings from above, but muppets are controlled with sticks from below. Just ask Jim Henson.
LLMs respond best to concise, well-defined, and well-constrained problems. If anybody working in the technology industry has a job solving these kinds of problems please raise your hand.
A good article. It's good to remember that any jobs eliminated by automation pale in comparison to those destroyed by mismanagement and poor leadership.
Mismanagement and poor leadership, part of which includes rejecting, misunderstanding or misapplying automation.
In the case of LLMs, though, I wouldn't fault any business leaders for being highly suspicious right now. It's a tech worth keeping an eye on, but it's not at all clear it will actually be that useful in practice.