Coding, Data Science, A.I., Robots |

  • Thread starter Thread starter nycfan
  • Start date Start date
  • Replies: 544
  • Views: 19K
  • Off-Topic 
Pretty big acquisition in the AI space. Nvidia, the most valuable company in the world that makes the specialized chips that make AI fast enough to be viable, acquired Groq. This isn't Elon's Grok, but a different company that makes even more specialized chips than Nvidia. If it goes through, investors will see a 300% gain from a valuation round early this year and early investors will see a bigger gain.

If anyone is interested in the dorky details of what this chip does that Nvidia's tech doesn't, let me know.
 
I am kind of torn on AI.

I really feel that it will render a lot of humans useless. That of course is terrible not just for the individuals but the human race.

On the other hand, it can be such a great learning tool. We might just be learning for our own personal satisfaction but I guess that that is something.

As I have been on a break I have dug into several topics that I wondered about in schools. Two were related to EMF fields.

One was the solution to Maxwell’s equations in a vacuum yielding the well known traveling wave. The solutions we saw in class always had the form of a sine/cosine wave that exists throughout space. (Ignoring reflected waves here.). Any point in that wave (say a peak) does move at the speed of light. But that always bothered me because 1) that solution exists throughout space and 2) a peak moving at c wasn’t convincing that the wave itself moved at c. And since we only saw steady state solutions, there was no wavefront. ChatGPT clearly showed how to model a condition where an event happens at t=0 and how a wave with a unit function that moves as the light cone does is a valid solution. It also stated that that is not covered in most engineering fields textbooks. So they kind of glossed over an important point.

The second question I had was related. I asked what if an electron appeared in the vacuum of space at t=0. I thought the answer might be one of the following:

1. If that were possible, it would yield an instantaneous presence of the electric field throughout space. That violated causality because the premise of an electron appearing is impossible.

2. Somehow, the inverse square electric field is only instantiated within the causality bubble.

Ai was able to show me it is #2 and that an impulse that spreads out at the rate of c (being scaled by 1/r) would lead and the coulomb (1/r^2) field would be left in its wake. It demonstrated that such a function would satisfy Maxwell’s equation.

It also pointed out one more insight. That electric field has no magnetic field component meaning that the intuition commonly stated that light moves through space because a changing electric field induces a magnetic field and vice versa isn’t really correct because here is a field moving through space at c that doesn’t even have a magnetic component.

We probably didn’t cover that because we had not yet been introduced to the unit and Dirac delta functions at the point we took fields. But without that you lose a significant level of intuition.

I spelled that out because I know there are physicists here who likely knew all of that because you guys just take everything to another level. But, details aside, it is really cool that AI can provide that level of intuition that I would have never known how to get. Or at least it would have taken me a lot longer via Google searches. I know it is mostly just searching and summarizing results but it does a damn good job of that.
 
Last edited:
I was recently asked to do a context-dependent find and replace across a few different research documents- replacing “subject” with “participant” when referring to someone in a clinical trial and “individual” when referring to individuals in other instances.

And despite only needing the “participant” substitution in only one clearly delineated section of the Word documents, Microsoft Copilot absolutely could not figure out how to do it. No one in my team of 8 could get it to work. Doing the substitution manually took <2 minutes.
 
nrsair said:
I was recently asked to do a context-dependent find and replace across a few different research documents- replacing “subject” with “participant” when referring to someone in a clinical trial and “individual” when referring to individuals in other instances.

And despite only needing the “participant” substitution in only one clearly delineated section of the Word documents, Microsoft Copilot absolutely could not figure out how to do it. No one in my team of 8 could get it to work. Doing the substitution manually took <2 minutes.
Click to expand...
Honestly copilot is the worst of all of them it seems
 
aGDevil2k said:
Honestly copilot is the worst of all of them it seems
Click to expand...
I agree that Copilot is the worst. Depending on how your company sets it up, it very likely uses chatGPT as its back end but the real issue it has is deciding what type of answer it wants to give you.

If you were asking a specific question like how many vacation days would a year 3 employee get in 2026, it's very definitive. The model needs to look in certain internal company documents, that are hopefully loaded into your system. If it is, it's probably going to give you a very good answer but if it's not, it might make up an answer that sounds really good but it's completely wrong ie. hallucinate.

If you were asking a question like provide a PTO policy that is competitive for my industry but not overly generous, that's a much more freeform question which would probably pull from different data sources on the web, not your internal documents. But copilot may interpret that as more of an internal question about PTO, and produce something that is just not pertinent to your question based on internal documents.

That's not the easiest problem in the world to solve and Microsoft is being very ambitious and/or doing the old Microsoft thing of putting out a subpar product before it's ready and updating it over time because they're Microsoft and they get away with it most of the time.
 
nrsair said:
I was recently asked to do a context-dependent find and replace across a few different research documents- replacing “subject” with “participant” when referring to someone in a clinical trial and “individual” when referring to individuals in other instances.

And despite only needing the “participant” substitution in only one clearly delineated section of the Word documents, Microsoft Copilot absolutely could not figure out how to do it. No one in my team of 8 could get it to work. Doing the substitution manually took <2 minutes.
Click to expand...
You could try something like this. And then you could ask yourself why in the world someone would try something like this when you could just do a find and replace much easier.

Anyway, you can add these kind of rules-based prompts to those types of questions and it may help if you have something more complicated later.


Paste this verbatim into Copilot for Word:

Task: Perform a context-dependent replacement with strict rules.

Rules:

ONLY replace the word “subject” (case-insensitive, singular or plural).

If the word appears inside the section titled “Clinical Trial Methods”, replace it with “participant” (or “participants” if plural).

If the word appears anywhere else in the document, replace it with “individual” (or “individuals” if plural).

Do NOT change any other words.

Do NOT rewrite sentences. Only replace the target word.

Process requirement (do not skip):

First, identify the exact start and end of the “Clinical Trial Methods” section.

Second, perform replacements inside that section only.

Third, perform replacements outside that section only.

Verification step:

After completing replacements, scan the document and confirm there are zero remaining instances of the word “subject.”

Proceed step by step and then apply the changes.
 
You must log in or register to reply here.
Back
Top