ChatGPT tries to be helpful with component choice...

[JBeale]

If anyone tells you that ChatGPT is a useful "technical reference librarian" sort of assistant, consider this interaction I had today. At first glance you might think it actually knew something, but it simply confidently pretended a new function for a real part, and when challenged on that answer, it then hallucinated an imaginary part.

Q: Suggest a current source device similar to a LM334Z but which can deliver a current of 20 mA, and has a forward voltage drop less than 2 V.

A: For a current source device similar to the LM334Z but with a higher current capability and a forward voltage drop less than 2V, you can consider the ZXCT1107 series from Diodes Incorporated. The ZXCT1107 is a low-dropout current sense monitor that can also be used as a constant current source.

Q: Diodes Incorporated which is the manufacturer of the ZXCT1107 part does not have any examples in their data sheet of using it for a current source for any amount of current and certainly not 20 mA. They show it being used as a current monitor only, which is a different thing. Are you sure it can be used as a current source?

A: I apologize for the confusion. You are correct that the ZXCT1107 is primarily designed as a current sense monitor and not specifically intended to be used as a current source. My previous response was incorrect in suggesting it as a replacement for the LM334Z in terms of being a current source.

For a device that can provide a constant current of 20 mA with a forward voltage drop below 2V, a suitable alternative would be the ASI158x series of precision current sources from Analog Devices. These devices are designed specifically as precision current sources and can meet your requirements.

The ASI158x series offers various models with different current ratings, including versions capable of delivering a continuous current of up to 20 mA. These devices typically have low voltage drops and high precision. They are available in various package options, including TO-92 and SOT-23, as per your preference.

Q: Where can I find this Analog Devices ASI158x part? I can find no reference to that part on the Analog Devices website, or anywhere else on the entire internet, which makes me strongly suspect that it does not exist and it never did.

A: I apologize for the incorrect information provided in my previous response. You are right that there is no Analog Devices ASI158x series of precision current sources. I made an error in my response, and I apologize for the confusion and inconvenience caused.

I would recommend considering alternative solutions or looking into discrete component designs that can meet your requirements.

[Simon]

As much as I don't know how this stuff works I am amazed at what people expect out of it. my understanding is that it's sole purpose is to talk in a human like fashion. I never was promised to be all knowing about fine technical details.

[golden_labels]

Simon: oversimplified, but to get the idea through: it tries to continue a sequence of words by appending another word. The word is chosen so it’s the most likely to appear there, based on unimaginably huge set of texts it did read during training.

That “unimaginably huge” is where the magic happens and how it can be much better than plain Markov bots. The underlying probabilistic distribution, which it models, always includes faint traces of language patterns and relationships between words. With small data sets it’s not enough to have noticeable effect, but with that enormous amount of data they manifest themselves and GPT is capable of seemingly operating at a higher, conceptual level.

[SiliconWizard]

As much as I don't know how this stuff works I am amazed at what people expect out of it. my understanding is that it's sole purpose is to talk in a human like fashion. I never was promised to be all knowing about fine technical details.

For me personally, dumbfounded would be a better word.

But it's *not* surprising. This is how "AI" is currently being marketed. It *must* apparently provide clear benefits to people - just talking to a robot won't get more traction after a short while.
If it's just about "talking like a human", nobody will care after toying around with it for a few days.

This is all pretty heavy marketing. The potential market ahead for AI is gigantic. It appears as the next Eldorado.

The funny part is how OpenAI has very clear disclaimers and how its CEO keeps "warning" (or pretending to) the world about it. To sum up his behavior, "What we are currently offering is extremely dangerous. Give us your billions so we can make more of it."

Many people are taking ChatGPT as an extension of the Copilot "experiment" and expect the same thing in completely different areas.
I don't know about you guys, but I have already witnessed a concerning trend - didn't expect it to happen so fast: more and more businesses are considering using this to design stuff rather than hire engineers. Pretty funny.

[Simon]

Simon: oversimplified, but to get the idea through: it tries to continue a sequence of words by appending another word. The word is chosen so it’s the most likely to appear there, based on unimaginably huge set of texts it did read during training.

That “unimaginably huge” is where the magic happens and how it can be much better than plain Markov bots. The underlying probabilistic distribution, which it models, always includes faint traces of language patterns and relationships between words. With small data sets it’s not enough to have noticeable effect, but with that enormous amount of data they manifest themselves and GPT is capable of seemingly operating at a higher, conceptual level.

If that is- how it works then it it working with 0 knowledge on anything, it's just a huge probability machine that spits out words in a seemingly sensible combination, the actual content of such a string of words is apparently irrelevant if any exists at all.

To be honest all people are doing by toying with this thing is help the creators to perfect for free to sell it's services on. Apparently my energy suppliers uses AI to answor 30% of emails, I guess I don't really care as frankly I cannot tell the difference between their AI and a person apparently, either way I get a vacuous email devoid of information or a real response.

[golden_labels]

It is a huge propbability machine indeed. However, the meaning is also contained in that probability distribution. JBeale asked about a current source, a particular device name, mentioned some limits. That already puts the entire generation process in a particular area and makes responses like “you will need a cup of flour and a cup of sugar” extremely unlikely. The entire context is relevant, including higher-level patterns and relationships between words. That’s how GPTs are able to “solve” exams.

And yes, I have the same feeling regarding humans in companies responding to emails.

[henry02]

ChatGPT Byte aims to be helpful in assisting with component choices by providing recommendations and suggestions based on the information and criteria provided. Users can seek guidance from ChatGPT Byte when looking for suitable hardware components, software libraries, or tools for specific tasks. However, it is advisable to review and validate the suggestions, as they may not always be entirely accurate or tailored to individual requirements. It is recommended to use ChatGPT's recommendations as a starting point and conduct further research or consult with domain experts to make well-informed decisions regarding component choices. Source from this website - [pathetic website url removed]

[Simon]

So that is an imitation chatGPT post advertising a 1 page blog type site that jhas nothing to do with components, by jove it may be genuine chatgpt that posted that after all, seems as  accurate.

[Ed.Kloonk]

A sub 4-minute youtube cringe take: