[alancalverd]

Ask ChatGPT to create some HTML code for a standard log in page of a website that your building and you'll get similar instantly useable results.

The interesting question is whether it is writing from scratch or searching for examples of the requested code. If the latter, you still need to check for any clash of variables, like using x to mean two different things or calling for a variable that doesn't exist, when you embed the new patch in something bigger.

This might explain why Companies House and BT email  are now full of bugs and the County Court claim filing system is incompatible with Explorer: updates written by an unaccountable (but very expensive, naturally) idiot! Or Fujitsu.

[hamdani yusuf]

The interesting question is whether it is writing from scratch or searching for examples of the requested code. If the latter, you still need to check for any clash of variables, like using x to mean two different things or calling for a variable that doesn't exist, when you embed the new patch in something bigger.

You can simply ask it. You can also ask it to check.

[Zer0]

Copyrights & Credits -
BBC News/Youtube.

ps - If a Picture says a Thousand words...
Maybe a Video could say More.

[hamdani yusuf]

Solar power is the most economic power source. Any massive usage of energy will make it the first option, in order to be more competitive. Other power sources are likely to be the back up, and will be used only when the primary power source fails.

[Zer0]

& perhaps whether We even should power it On, on such massive scales, or give it more Time for refinement inorder for IT to get things Right is also an Option which could be taken into consideration.

The BBC Guest seemed to be mildly opinionated about AI.

But She did say something quite important & necessary...
" Trust = T.E.A. "
(Trustability Explainability Accountability)

T

https://www.theguardian.com/commentisfree/2024/feb/13/software-student-cheated-combat-ai

(it creates trust issues)

E

https://theconversation.com/why-humans-cant-trust-ai-you-dont-know-how-it-works-what-its-going-to-do-or-whether-itll-serve-your-interests-213115

(a bit too many unknowns)

A

https://www.forbes.com/sites/marisagarcia/2024/02/19/what-air-canada-lost-in-remarkable-lying-ai-chatbot-case/

(the law which is the law is the law)

ps - enough words, now time to wait & watch, how the future unfolds.
: )

[hamdani yusuf]

As a seasoned ChatGPT user, you're familiar with the insights and assistance a single Language Model can provide. But what if you could harness the collective capabilities of SIX different Large Language Models at once?

Check Out Chathub and Change the Way you Use LLMs!

[alancalverd]

As a seasoned ChatGPT user, you're familiar with the insights and assistance a single Language Model can provide

So far, no evidence of insight, just uncritical mashups of stuff that's already out there.

[hamdani yusuf]

As a seasoned ChatGPT user, you're familiar with the insights and assistance a single Language Model can provide

So far, no evidence of insight, just uncritical mashups of stuff that's already out there.

Tell that to those who've just got laid off from their jobs, related to increasing use of AI.

[alancalverd]

Nothing new there.

25 years ago one of my children's college friends got a summer vacation job with an airline freight handler, initially entering data into a spreadsheet.  He quickly realised that with a few extra command lines he could get the spreadsheet to do all the rest of his work, so he sat at his desk for a few weeks watching the computer earn his wages. Then it occurred to him that he could replace everyone else in the office with a bit more software. Moral problem: tell the boss, get promoted and make 5 other guys redundant, or stay stumm and come back at Christmas?

[hamdani yusuf]

However you see it, the machines seem to have growth rate much higher than their biological counterparts.

Anthropic just dropped Claude 3, a cutting-edge model that performs better than GPT4 across the board, according to benchmarks. I'll tell you all about it, and then we'll test it ourselves!

Chapters:
0:00​ - About Claude 3
8:35​ - Pricing & Use Cases
10:47​ - Testing

[hamdani yusuf]

But what is a GPT? Visual intro to Transformers | Deep learning, chapter 5

Timestamps

0:00 - Predict, sample, repeat
3:03 - Inside a transformer
6:36 - Chapter layout
7:20 - The premise of Deep Learning
12:27 - Word embeddings
18:25 - Embeddings beyond words
20:22 - Unembedding
22:22 - Softmax with temperature
26:03 - Up next

[hamdani yusuf]

Coding a ChatGPT Like Transformer From Scratch in PyTorch

In this StatQuest we walk through the code required to code your own ChatGPT like Transformer in PyTorch and we do it one step at a time, with every little detail clearly explained.

0:00 Awesome song and introduction
1:12 Loading the modules
2:04 Creating the training dataset
6:17 Coding Position Encoding
14:09 Coding Attention
21:04 Coding a Decoder-Only Transformer
26:39 Running the model (untrained)
29:18 Training and using the model

It's said that the ability to create is the highest level of thinking.

[alancalverd]

Said by those who create, presumably. But create what, and why? The higher form of thinking is to establish a clear need and specification - nailing the bits together doesn't involve judgement or visualising context.

[varsigma]

What I think we might eventually see is an AI able to 'create' programs that it can prove are correct.

Why that is is because humans can do it already and all you need is logic. Induction and whatnot.
So to me, that will be down to being able to train some AI to use predicate logic and induction on some invariants, of loops and branches in the algorithms, however these are presented.

The faculty had cobbled together a "program logic" and embedded it as Java in C⁺⁺, as "comments" (erm, yeah another course I did). But an AI would in a sense, be that embedded code.

[hamdani yusuf]

Said by those who create, presumably. But create what, and why? The higher form of thinking is to establish a clear need and specification - nailing the bits together doesn't involve judgement or visualising context.

It's Bloom's Taxonomy.

[alancalverd]

At some point, future AI models will be able to rewrite its own source code.

Thus rendering the bot legally liable for whatever it says or does. But does it have any money to pay damages? If not, then the liability will lie with whoever deployed it. Users beware - you may be liable for unlimited damages incurred by a machine over which you have no control.

A sound legal precedent, where a manufacturer of a defective product went bankrupt, has famously led to the prohibition of using any Dupont material in a medical device.

[hamdani yusuf]

At some point, future AI models will be able to rewrite its own source code.

Thus rendering the bot legally liable for whatever it says or does. But does it have any money to pay damages? If not, then the liability will lie with whoever deployed it. Users beware - you may be liable for unlimited damages incurred by a machine over which you have no control.

A sound legal precedent, where a manufacturer of a defective product went bankrupt, has famously led to the prohibition of using any Dupont material in a medical device.

Don't confuse ability with permission. Many people are able to drive cars while not having license, or not having license to drive in some certain area.
Many people are able to kill someone. It doesn't mean that they're allowed to.

Manufacturers are expected to meet users' requirements. When the system affects wider society, regulations are needed to set the standard for minimum requirements.

[alancalverd]

The required regulation here is that the originator of any AI system is liable for any unintended consequence of its deployment if the end user does not have the resources to compensate victims, or cannot reasonably be held to have prior knowledge of the likelihood of such consequences.

There's plenty of precedent: it is illegal to drive a car without third party insurance, and the manufacturer is liable for any inherent fault that causes harm. The only glaring oddity is the constitutional right of third-world citizens to kill each other with guns.

[paul cotter]

A couple of very minor disagreements, otherwise i'm in full agreement. (1) In general motor vehicle insurance is obligatory with one notable exception: in New Zealand there are alternatives to insurance but it is still recommended. (2) In the US it seems that murder via firearms could soon become state sanctioned- shoot all shoplifters!!

[alancalverd]

And of course the US President is now exonerated from all criminal charges when in office.