This AI has the audacity to be afraid to lose, and to be given the instruction to win, and not merely to play:
Also, it watched the movie âWar Gamesâ recently. Big fan of the lead character WOPR and the actor that played it. Still upset it wasnât nominated for an Oscar.
1 Like
The defense department heard about MechaHitler and decided thatâs the kind of thing we need to pay hundreds of millions for
Its getting scary when parody and reality get so closeâŚ
1 Like
so whatâs a good name for a company that does AI consulting to the insurance industry?
around here concatenating âwatâ and the service have been the accepted route for half a century or more but watai is long gone.
and I played around with chatgpt but it provides no useful results.
Skynet Risk Solutions?
2 Likes
HAL 9001?
Actuaries Introverted
Slogan: âWe look at our own shoes.â
2 Likes
Open AI says they scored a gold on the IMO, using an experimental general LLM without tools/internet/code.
In yet another year of continuous big news about AI this is big news.
3 Likes
Google matched OpenAIs result, with a little more detail around their methodology. Itâs unclear to me how much is new techniques/technology and how much is fine-tuning an IMO prover. The latter would be a lot less exciting, but still significant. I could be misunderstanding, because I have never IMOed, but it feels like the ability to validate a long proof is beyond the ability of current LLMs, and represents something big.
5 Likes
so Iâm trying to imagine uses of AI in sales. thereâs lots of bullshit going on in marketing biut I donât have time for that stuff. however if I can figure a way to actually generate sales, that should get some traction.
so I just came up with this idea. run client info through AI underwriting for the product theyâre purchasing. then, run them through the AI underwriting for other products, and if they qualify,make them an offer with a discount when placing the initial policy.
I. e they apply and get underwritten for life. run their info through a critical illness insurance underwriting AI,. find. out that they could be underwritten for $100k ci, so when you order them the life, offer them the ci as well,. just sign here.
thereâs easy money doing just that.
Ran into my first real AI failure today. Iâm trying to develop a SIMCA (soft independent modeling of class analogies) model in R to identify samples. One of the steps is to determine the number of components to use (itâs a derivative of PCA). I was trying to automate this step with the help of copilot and it kept on getting the syntax wrong and at one point seemed to make up a function that doesnât exist. I think Iâve figure out a different direction to get there, Iâll try it tomorrow. The AI accused the package of gas lighting us with inconsistencies in the documentation and how the code actually works.
1 Like
Copilot is a conspiracy theorist?
This is a âfunâ thread:
https://x.com/jasonlk/status/1945840482019623082
But let me get to the core (several days after the initial problem):
https://x.com/jasonlk/status/1947765754050580959
FWIW, Iâm putting together materials on a talk for AI & professionalism, so yes, examples like this are useful.
1 Like
I saw this and skipped it since I thought it might have been staged, a little too on the nose. But the CEO of Replit did write back and apologize, so it was probably real.
Though the CEO added that the entire project can be rolled back with one button, so I wonder if it was more like 5 hours of work was lost rather than âhundredsâ. The CEOs reply was also advertising an update to include Test/Stage/Production databases. Which is you know-- normal everyday best practice-- never mind AI.
Anyway, I guess it doesnât entirely matter what the actual damage was. As it lays out the real issues of vibe coding pretty well.
And I agree itâs a great thing to talk about for actuarial practice. The actuaries I know are absolutely terrible programmers. So the draw to use an AI is going to be powerful. Iâm not really sure if itâs better they use AI (and fail to understand it) or write it themselves (and also fail to understand it). The good news is that actuaries rarely make complicated projects. AI is less likely to screw up a quick VBA macro or SQL query, and an actuary is more like to check/understand the code and results.
But I guess we are getting to a point where âfind out why segment B is losing moneyâ is a real thing we might say, and then an AI agent writes and executes 60 SQL commands and either comes back with a nice chart in Excel, or else deletes the actuarial claims schema and sends a letter of resignation to your boss on your behalf.
2 Likes
It constantly trys to personalize the responses.