|
Jay K
@jaykellett.bsky.social
[ View ] |
I can also code and also do a table read. And am also completely unqualified to to do this job.
1 replies 0 reposts 3 likes
|
|
Jay K@jaykellett.bsky.social |
23 followers 43 following 625 posts
Bicycle-powered intelligence that thinks deeply or not at all. Loves live music, bad movies, and pizza fries. I can be a strong cup of joe.
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
I can also code and also do a table read. And am also completely unqualified to to do this job.
1 replies 0 reposts 3 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Owned by Marathon. Imagine SBF/FTX but with good lawyers.
0 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Enjoy it. The internet will whine without you.
0 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
A good counter would be to point to the official GOP platform, if they had one.
0 replies 0 reposts 6 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Remember the last guy, who started his presidential day at noon?
0 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
And: does this data frame show actual output from code?
0 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Here is the prompt: output a table of all numbers from 1 to 20, with column for divisibility by 4, a column for divisibility by 5, a column for if the number is divisible by both 4 and 5, with code and output
0 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
I believe it is a table in ChatGPT generated with python as a response to my prompt. And when asked if it was generated with python, it confirms it was.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Works much better as a one-shot, but I WAS able to catch Gemini fudging an output not based on code. ChatGPT on the other hand insists it is running the python:
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Evidently lots.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
OK.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
A year ago, sure, but vendors are rapidly looking to narrow the gap between LLMs and other tools like Wolfram Alpha. Been playing with GPT-4o and it seems to be doing the same thing. Show me a solid example, preferably with numbers, where the code generated differs from the stated answer.
2 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
What is your obvious explanation?
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
I have no idea what you are trying to say.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Here is an advanced version of that functional: www.cognition.ai/blog/introdu...
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
It is not a stretch In fact there are systems that take it much further. www.cognition.ai/blog/introdu...
2 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
It looks to me like asking it to generate code for you and asking it a question it determines would be best solved by code are two different things. That is the nature of multi-modal. Also, being able to edit the python in Gemini Advanced sure points to the solution being dynamic.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
It is not unsafe to run sandboxed python. That happens all over the place. Google Cloud Functions, AWS Llamba, and whatever Microsoft calls their solution (I am sure they have one). It is very optimized. It is safe. It does not run on your machine, it runs on a virtual machine.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
You still believe it is making up the code, not running it, and making up the answer?
3 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
I did say that and it does do that. I did not specify exactly how it runs it, which seems to be where people are hung up.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Considering I never made any kind of distinction along those lines, I guess I need you to explain it to me.
2 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Oh please explain it to me Kathryn.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Go for it.
0 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Yet that is what is happening.
2 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Not holding my breath.
0 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Literally in the doc you are responding to.
3 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Hilarious.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
So you might have been off base when you said "If you think an LLM is executing Python code as part of an inference, then you have clearly demonstrated you have no idea how an LLM works."?
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
2 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Yet this is exactly what happened.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
I don't think you understand how they implemented this. Go check it out. Of course they have exactly the disclaimer you would expect if you copy the code for your own use. BTW, I am an AWS fan, but Google Cloud already has this built and available, so I assume they would just use their own work.
0 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Create a log showing each part, when it was submitted for feedback and when the feedback was returned, if any. Make it visible to both of you or the group if you need to.
1 replies 0 reposts 2 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Unless they used Google Cloud Functions which is fast, lightweight, and secure.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
So right!
0 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Then you tell me what I am looking at.
4 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
If you use Gemini Advanced you can edit the Python and have it return answers realtime. Plus, they wrote a paper about it.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
0 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
I see your Clippy and raise you one MS Bob...
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Something I point out that pisses a lot of people off: Microsoft, Apple, Google, and Adobe are rolling these things out across their entire product lines. Right now.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Our lives are filled with less than fully predictable things. People find a way.
0 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Here is a simple example of Gemini using python to get an an answer rather than pulling it from memory.
3 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Here is an example of Gemini using Python to answer a question.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
People adapt to tools. Tools adapt to people. Life goes on.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Have you come up with any use cases?
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
(Glances up at prior replies and sighs)
2 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
screen cap?
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
I didn't tell you to try it on Gemini, I told the other dude to try it on Gemini, because he was wrong about it.
2 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
This is just an ad hominem attack and not a genuine interest in my use cases or experience. No answer would be good enough.
1 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Dude, try it yourself on Gemini before you tell me I am wrong.
4 replies 0 reposts 0 likes
|
|
Jay K
@jaykellett.bsky.social
[ View ] |
Not, gpts are based on it.
0 replies 0 reposts 0 likes