Looks like a typical American highway, but-for the wrong font being used on the sign. 
1 Like
clearly merging pics from US and GB . . . or maybe Australia . . .
With spelling from the Chinese, no doubt.
John Murdoch has been looking at the impact of AI on jobs on the FT and has some interesting graphs.
AI is starting to affect jobs in a few select industries.
But as I suspected, the more judgment and complexity (messier) that a job entails, the less of a chance AI will have to supplant you. It is simply not that great at lateral thinking (multi-tasking several different issues at once in order to problem solve)
So if you have a crank the handle type job, I would be worried. Otherwise, not really.
Yeah. It can do less messy professional jobs no problem, but for long complicated professional jobs itâs only gone from 1% to 60% in less than 3 years⌠so uh⌠nothing to worry aboutâŚyeah.
Tried to use this internal version of Chatgpt to do some very basic data exploration and it couldnât do it. I think weâve got a few more years yet.
Periodically, I throw a âwrite me code that does [stuff] with data formatted [like this]â job to ChatGPT before writing something new for whatever the ad hoc task du jour is.
Iâve seen noticeable improvements in what itâs putting together for me. My most recent attempt was actually usable on the first prompt and saved me a bit of work.
(Although âusable on the first promptâ might as much a sign of my gaining skill in writing prompts, as improvements with ChatGPT.)
Decided to give Gemini 2.5 a test. I asked it to generate a traditional roguelike RPG inside of Excel, complete with generated little monsters that run around and attack you.
It instantly wrote a couple thousand lines of code across a dozen modules. Which I had to copy and paste. And ran into a few VBA specific errors, though nothing major. Pretty neat.
You can tell itâs real Excel game because my MP is 10-Oct.
I am going hard on ai. Ive got a âside developerâ working with me as we start to build a chatbot.
Ai is a bit weird to understand. Im getting there, partly by just ignoring the tech aspects of some things.
The two of us had a 2.5 hour meeting on the weekend to hammer through how to âloadâ data into an ai chatbot. In the standard chatbot building block model, you can only load blocks of text. SoâŚhow do you maintain the information included in a table? Or a pdf document with a heading, a bunch of text interspersed with tables, and some of the table cells containing ranges. You cant just read the text from the table in because then you lose the information included in column and row headings.
WaPo announces partnership with Open AI
Meanwhile:
Why hire DJs when you have AI?
Radio stations here hardly ever have DJs, just recorded BS between songs. You know what I want to hear? SONGS!! Your prerecorded banter is wasted time. Your 30 seconds of clips of songs you might play are wasted time. Your actual DJs who just talk to each other are wasted time.
That said, there is a great station, âKTYD,â in Santa Barbara that we listen to on the internet (and when weâre driving through to wine country) with actual DJs who sometimes seem to play whatever they want, even take requests on Friday Morning.
They have a Saturday Afternoon/Evening program (pre-recorded) called âThe Deep Endâ that plays not-on-the-500-song-list-that-every-classic-rock-station-that-has-to-make-a-profit-for-the-corporate-shareholders-must-play-from.
Based on the latest research done by Apple it does look to me that the majority of Actuarial work is going to be ok
The AI failed miserably at complex work and hallucinated responses.
This is why its just never going to replace a human in higher value add Actudonk work because the potential liabilities here would be enormous. No company would ever take that risk.
I do think itâs funny that they fail the river-crossing-puzzle. They can be so dumb and smart at the same time. I donât think thatâs a forever problem though. They could probably already solve the problems using code. Later they will have other solutions.
Thatâs aside from your thoughts on actuaries, which I agree are shielded from AI, for various reasons (money, regulation, spreadsheet format, visual reasoning, legal reasoning, etc.) I do think specific pieces of our jobs will be vulnerable though.
1 Like
It will be very interesting. The way I think of our professional development process is learning the basics helps form the actuarial understanding at a big picture level. If we skip/automate away all the grunt work, what will we be losing in formation?
3 Likes
I think current AI just doesnât actually do any thinking, it just wants you to believe it has. I asked chatgpt to build me a schedule to train for a marathon, which it did a great job of. I imagine it digested plenty of others created on the internet so could do a solid job.
Then I asked it to give me appropriate runs starting from my address that fit the schedule it provided me. It responds like it understands the question but proceeds to completely make up streets as if they intersect and was utter garbage.
3 Likes
So, it should be called âA.B.S.â?
As best I can understand, the training data is all put into vectors with thousands of dimensions. Given enough vectors, you end up with a model which is just a space containing those vectors.
Then when it needs a response it just looks at the associated vector and thereâs your answer.
THe one example I saw was taking your photo, then aging it. All the âAIâ did was slide the vector along one dimension that represented age. Boom, new picture.
Thatâs the âthinkingâ - just creating a model and then defining a vector to pull out of the model.
Theyâve done it with images and video and language, and other things are coming online fast. Customer service is already a dead industry - people are losing their jobs right now. Say goodbye to your friends in the call center at head office. Underwriting? Thatâs going away starting this year.
âActuarialâ is pretty broad, but thereâs certainly a lot of functions that are going to go AI. Pricing, stuff like that. Just need enough training/data points to create the âuniverseâ and then everything follows. Probably anything mathematical will be toast, including âjudgementâ calls.
But I think all that means is that actuaries in some niches will end up doing stuff a bit differently.
2 Likes
It is a forever problem in this way.
The models cannot learn and dynamically apply symbolic rules. This is how we deal with novel situations, from solving an addition problem we have never seen, to performing medical triage after a major disaster.
We seem to weave together pattern recognition and different systems of symbolic rules in a very complicated way. That, too, is missing from the models.
This still leaves out our ability to have goals and values, our theory of mind (the ability to infer what other people are thinking or want), and finally our embodied existence, all of which are also missing from current models.
Personally, I donât know how anybody thinks this current round of technology will give us AGI.
1 Like