this was totally worth the time investment, recommended:

Thinking about getting a new keyboard. Haven't had Cherry MX switches before, so I'm not sure if I want Red or Brown. I currently use a GoldTouch split keyboard that has a fairly springy feel and is not terribly loud. I want another split keyboard (preferably one that allows tenting to some extent for ergonomics). Any thoughts or suggestions welcome.

That princess bride video is amazing. Hugh Jackman in a bathrobe is killing me. 😂

Watched some of the new chatGPT 4 videos. We'll all be obsolete in about 36 months methinks 😉

But seriously, is someone on the game of adding all Lucee / Commandbox / CFML documentation to it?

@jakobward has left the channel

its very strong i would say for writing passable code for algorithms where there are many documentation examples. Asking it to write an implementation of an algo im working on i found that it wrote passable code for CFML. It understood for example that arrays in CFML start at 1. it lies routinely or has extreme confidence which is something to watch for so for example it relied in on a BIF called ArrayShuffle() when i pointed out no such function existed it provided a replacement which was solid. Then i had to remove when var from one line of code where it was calling outside a function and ultmately you can't local scope a call in the way it was trying but very solid

I would say its GO is better than its CFML but even then it would try and use INT64 in places where a big int was more appropriate if you told it this code is going to overflow it would then rewrite with big int throughout

it doesn't write tests though., routinely I'm interested in widespread ingestion of our code reviews and coding standards such that you could train a model to provide more immediate feedback

I was impressed that when i fed it a generated sample of CFML and said what could be wrong with this from a security standpoint it identified parametisation of queries

So its a bit of a liar but definitely impressive

I'm using it pretty routinely with Google gpt sheet extension and i use it a lot of the time for looking up legislation or rules around certain key points

lawyers, accountants or any jobs based on patterns or rules are F&**Ked imho

Deepl.com has already ruined it for translators

In real life IT we have much trouble with fuzzy business requirements, conflicting rules, incompetence of people making decisions, sometimes with direct lies -- things that only human can comprehend. Great progress for GPT model, no doubt, but it's relatively simple to write code solving some well-defined problem, solved thousand times before, like students do on studies, eg. writing numerical methods solving for diff-equations (first thing came to mind). The other, harder thing, is to have some inventive, creative approach to uncertainty of real life.

Agreed but the models have leaped in the last 12-24 months noticably. And GPT4 is another leap against previous model.

Code creation is a utility is helpful, in the same way as a spell check on steroids as you type

limited utility, where it becomes i would suggest a game changer is its ability to infer outcomes

I look forward to everyone being unemployed and looking for things to do as Skynet kills us all

for example In the membership association world, there is a high degree of commonality between those that engage (consume services) and those that retain membership. This is relatively obvious. If you feed data around signals, lack of signals and profile of users being able to infer that someone based on their direction of travel is unlikely to renew thats valuable

now of course take an experienced membership manager who knows the members they might be able to make the same determination but that doesn't scale

I think where knowledge is applied to context and learned patterns is where it offers some interesting stuff

I kinda like the idea of a technical concierge that knows data and surfaces it to employess to say oi this person looks at risk wanna check.

But some f&**ker is bound to use it for some evil shit

I look at it and the first thing I think of is how can I break it and how can I use this in the worst way possible. For example, in the US almost all contact and transactional and personal data is for sale, legally via credit card companies and banks. Using that with an Ai, to process a persons eating, purchasing, and personal habbits, one can with current tech "GUESSS" with a high level of accuracy the voting demographic the persons belongs in, and what shiny new thing they might purchase next. Now I will couple an AI with ads and "SMART" coupons, so that I know I can sale any item at a certain point with a high degree of probility. IE THey will buy a box of cereal at 3.99 but not at 3:29

my teeth itch.

Hopefully nobody will think of anything worse that using AI to make you buy cereal 🙂

AI is hugely dependent on the training material and then training material is going to be largely human generated historical data with all it's in-built conscious and unconscious biases.

And as Alex has said, the context of a sentence makes a huge difference so can be just as important as the actual sentence.