How ChatGPT can help push universities to change

[quovadis]

https://en.wikipedia.org/wiki/Large_language_model


They are literally just a really powerful auto-predict.

If you want to oversimplify it and ignore the differences in GPT's implementation vs previous generation LLM's then perhaps.

 

[konfab]

If you want to oversimplify it and ignore the differences in GPT's implementation vs previous generation LLM's then perhaps.

What is the task that ChatGPT is trained to do?

 

[quovadis]

What is the task that ChatGPT is trained to do?

Have you experimented with it? Tell it to be a SQL server, then import data and run queries against it. While obviously not efficient use of processing power - that's far more than being simply characterised as a powerful auto-predict. I haven't touched 4.5 as yet but with 4 I've been able to have it reproduce code based on specification alone which results in source code arguably just as good as any dev but in the time a dev would probably take to read the specification's title.

 

[konfab]

Have you experimented with it? Tell it to be a SQL server, then import data and run queries against it. While obviously not efficient use of processing power - that's far more than being simply characterised as a powerful auto-predict. I haven't touched 4.5 as yet but with 4 I've been able to have it reproduce code based on specification alone which results in source code arguably just as good as any dev but in the time a dev would probably take to read the specification's title.

I have not stated it isn't an exceptionally powerful tool. I have already used it to generate code that I have deployed into production.
If you can specify all your requirements, it does a very good job with code generation.

Nevertheless, we are ultimately just using emergent behaviour that results from this simple task.

 

[quovadis]

I have not stated it isn't an exceptionally powerful tool. I have already used it to generate code that I have deployed into production.
If you can specify all your requirements, it does a very good job with code generation.

Problem is that for most problems, you cannot really specify the requirements in anything less than the code.

Please don't get me wrong though - my criticism was solely on the simplified characterisation. I've found that the contextual understanding of the "chat" though is impressive enough to overcome some of it and 4.5 is stated to be much better so we'll see.

The ramifications of this technology is going to be scarey.

 

[konfab]

Please don't get me wrong though - my criticism was solely on the simplified characterisation. I've found that the contextual understanding of the "chat" though is impressive enough to overcome some of it and 4.5 is stated to be much better so we'll see.z

The characteristic of an LLM as a glorified auto-predict is very correct IMO. It is literally how it works.

The ramifications of this technology is going to be scarey.

I don't think so. It is good if you can specify exactly what you want, but that is most of the difficulty with most things.

 

[quovadis]

The characteristic of an LLM as a glorified auto-predict is very correct IMO. It is literally how it works.

I think we're missing each other in meaning. I'm not disagreeing on its functional model behind GPT.

I don't think so. It is good if you can specify exactly what you want, but that is most of the difficulty with most things.

You don't see an impact to employment in terms of automation of tasks, less entry level/repetitive jobs etc?

 

[FiestaST]

Learn something new everyday.