VectorLock

I wouldn't exactly call this suicidal ideation, but maybe a topic to broach at your next session.

If you promote it less you decrease its "acceleration" dangers, I guess.

Nobody complained to Farnsworth.

You've been able to do that for a while, just depends on how large scale you want to get. Been doing it since the 60s.

I think GPT is probably best at generating Python code, as well.

All you need to know is the volume of the jar and the packing fraction of jelly beans.

I won a contest in 5th grade science class with that info.

Almost as interesting as the GPT part of it.

One of the 15 or so "risks" that OpenAI supposedly tested for[1], below things like "Hallucinations" and "Weapons, conventional and unconventional" was "Acceleration."

I thought this was a really interesting topic for them to cover. In the section was 1 paragraph about how they're still working on it. Guess it wasn't, uh, much of a concern for them...

[1] https://cdn.openai.com/papers/gpt-4-system-card.pdf

When I was programming in DOS way back in the day Ralf Brown's Interrupt List was THE bible.

Especially for tax codes. Every tax break is a richer man's loophole.

Even harder when you've turned on Slack auto-delete for "security reasons."

Really Slack is more approaching IRC when configured like that, except I could still scroll back my IRC client. In Slack they're just _gone_.

>Flushing

When it gets really bad the person listing it will just flee the country, for sure. Assuming its not already straw man'd out.

Was there as much handwringing about computers and databases putting secretaries out of work?

Changebacks aren't guaranteed that your bank or credit card issuers will allow it. I've had a bank tell me several times that I can't do a chargeback when there were fraudulent purchases made on my account on the Playstation Network.

Even better, Sony threatened to ban my account and I'd lose all the content I legitimately purchased if I did a chargeback against them.

I buy physical game copies now and I've found CapitalOne is better at doing chargebacks than my credit union is (guess credit unions aren't all they're cracked up to be in certain aspects)

I know the answer. SNS can't do that. But ChatGPT hallucinated it could. Just like the original post about a capability their API doesn't provide.

The #1 reason everybody gives for "RTO" is "to improve collaboration." It seems like having hot desks if implemented properly could MAYBE actually help facilitate that. Let teams get areas (or, gasp, dedicated OFFICES) where they can sit together when they coordinate for in-office collaboration time. Let people move around to shoulder surf or pair code when they want to. Having hot desks with a hybrid office model makes much more sense rather than having people go in to a 3/5ths full office and try to ignore the distractions.

If you want to put headphones in, nose down, and get into flow that is the beauty of WFH. Unless you have a REALLY unconducive home environment to that - which I know some who do but the majority have very well optimized home setups that allow them to focus without distraction.

This is the biggest problem I encounter when trying to use ChatGPT on a daily basis for computer programming tasks. It "hallucinates" plausible looking code that never existed or would never work, especially confusing whats in one module or API for something in another. This is where ChatGPT breaks when pushed a bit further than "make customized StackOverflow snippets."

For example I asked ChatGPT to show me how to use an AWS SDK "waiter" to wait on a notification on an SNS topic. It showed me code that looked right, but was confusing functions in the SQS library for those that would do the thing with SNS (but SNS doesn't support what I wanted)

Secretary was the most populous job in the US before the advent of personal computers.

How were you getting the report from ChatGPT into the form? Sounds like you'd have to tab/F2 anyways to input what ChatGPT says.

I think a vanishingly small number of criminals have police scanners and radar detectors in this day and age. Especially since police have largely moved over to encrypted trunked radios.

Definitely marketing hyberbole but if they need to use the tag dart, presumably it is someone who is fleeing from the police, so they are _at least_ guilty of eluding, with probably a very high conviction rate for at least that offense, if they go to trial.

"Anybody who cares about linting rules shouldn't be picking the linting rules." - Vector's Razor

I have also asked it to generated AWS related code in Python and it has something wrong every single time.

Its incrementally better than just copy & pasting from StackOverflow, since it will customize the code for you, but if you try to go beyond what can easily be found on StackOverflow it will fail you.

This could be interpreted as an indictment of ChatGPT or the article's author.

Maybe ChatGPT can finally drive us over the edge to reset those algorithms.

The article was so long and rambling I asked ChatGPT to summarize it for me:

The increased use of generative AI tools like ChatGPT has brought to a head the problem that creators have been facing for some time. They worry that these tools are greater masters of their craft than they are. According to a recent article, the explanation is simple: creators have overemphasized one part of the craft, posture. This refers to how creators see themselves and the world, which creates a unique creative “fingerprint”. Creators are not focusing enough on the other two parts, the process, which is how the work is directed, and practice, which involves honing creative skills. Instead, they are relying on techniques, workflows, and tools to direct them. By learning and focusing on the process and practice, creators can improve their skills and not worry about being replaced by AI tools.

Now going back to the original premise, despite this code being wrong, as ChatGPT almost invariably does, do you think it understands the concepts here or is just statistically generating tokens based on previous input?

Really for code generation ChatGPT is an incremental step over StackOverflow. It can kind of nicely template the things you tell it into the code, for the most part, but the code is almost always fundamentally wrong or just mashed together in some way.

I haven't tried Bing but ask ChatGPT "what is 12348120348102341 + 10823401828482?"

Bonus points: If you ask Google Search this, it knows the right answer.

If you don't think its just statistically choosing the next word, what do you think its doing? Do you think it understands the mathematical concepts you're questioning it about?

What the article describes is not that. They're saying its an extra shrinkage ON TOP of the usual 1,000,000,000 bytes = 1Gb thing drive makers have already done for a long time.