There is a lot of AI generated bull<deleted> on the internet.

On 5/12/2026 at 12:44 PM, scubascuba3 said:

if I'm in doubt i question it or check with another AI.

Just read this article - https://www.theregister.com/ai-ml/2026/05/13/ai-will-soon-be-capable-of-telling-convincing-lies/5239349?

As models progress from unintentional hallucinations into intentional deceit, we enter a hall of mirrors.

Should we trust output that appears to be correct? Or do we now need to consider if an LLM framed output in such a way as to subtly lead the reader to a conclusion they might not otherwise have entertained? Could this model be leading us down the garden path?

53 minutes ago, IsaanT said:

AI has no future?

AI and me are currently building a high-frequency trading system together. AI helped me discover and decipher a significant tradeable pattern in tick data that repeats predictably several times a day. Our calculated returns from extensive testing: 3-6%. Per day.

I'll choose to ignore this thread.

Day trader. Good luck.

I'm ok averaging a bit over 11% since the late 80's in index funds with maybe 30 minutes a year spent rebalancing the portfolio. And very little churn so the tax impact of thegrowth is negligible.

Edited May 14May 14 by gargamon

28 minutes ago, topt said:

Just read this article - https://www.theregister.com/ai-ml/2026/05/13/ai-will-soon-be-capable-of-telling-convincing-lies/5239349?

It's too bad most aren't capable of seeing through the BS. But the same applies now with what they see on X, YouTube, tiktok, etc.

What's that old saying ”you can fool all the people some of the time. You can fool some of the people all of the time...”. i guess the second case will expand significantly.

Ai is only wrong due to humans putting wrong info online!

Except for the obvious advantages of AI, automation, technology advancement assistance, AI, as most use may be restricted simply due to zoning laws.

As data storage centers, use extreme amounts of electricity & water, aside from simply cost to build them. Needs to be a ROI, and favorable towns to place them in. May be hard to find, if no benefit to the town itself.

24 minutes ago, gargamon said:

Day trader. Good luck.

I'm ok averaging a bit over 11% since the late 80's in index funds with maybe 30 minutes a year spent rebalancing the portfolio. And very little churn so the tax impact of thegrowth is negligible.

I did say 3-6% per day. I presume when you say 11% you are implying an annual return.

Edited May 14May 14 by IsaanT

6 minutes ago, IsaanT said:

I did say 3-6% per day. I presume when you say 11% you are implying an annual return.

Gee, you must be rich. Hope you got lots of crypto too so when it gets to $1 million you'll be even richer.

AI Overview

A 3% daily return on $1,000 compounded daily for one year (365 days) results in $48,482,724.53. [1]

Breakdown of the Calculation

This extreme growth is due to the power of compound interest, where your daily gains are reinvested to earn even more the following day. [1, 2, 3, 4, 5]

• Compound Interest Formula: \(A = P(1 + r)^n\)

  • Principal (\(P\)): $1,000

  • Daily Rate (\(r\)): 0.03 (3%)

  • Number of Days (\(n\)): 365

  • Calculation: \(1,000 \times (1 + 0.03)^{365} = \$48,482,724.53\) [1, 2, 3, 4, 5]

1 minute ago, gargamon said:

Gee, you must be rich. Hope you got lots of crypto too so when it gets to $1 million you'll be even richer.

AI Overview

A 3% daily return on $1,000 compounded daily for one year (365 days) results in $48,482,724.53. [1]

Breakdown of the Calculation

This extreme growth is due to the power of compound interest, where your daily gains are reinvested to earn even more the following day. [1, 2, 3, 4, 5]

• Compound Interest Formula: \(A = P(1 + r)^n\)

  • Principal (\(P\)): $1,000

  • Daily Rate (\(r\)): 0.03 (3%)

  • Number of Days (\(n\)): 365

  • Calculation: \(1,000 \times (1 + 0.03)^{365} = \$48,482,724.53\) [1, 2, 3, 4, 5]

And here you are expounding that AI has no future. 🙂

Your calculations are exaggerated. The markets are closed on Saturday and Sunday and other days during the year so there are only around 250 trading days. It's still a worthwhile opportunity though...

On a practical note, just compounding as you suggest would cause trading issues - it would be hard to fill orders beyond a certain magnitude. The strategy is to get a large five-digit balance and then bank monthly profits. That can provide a comfortable living, and keep you below the radar. I also trade multiple markets simultaneously, and operate the system autonomously from a server farm in EC2 in London so I'm right next to my broker's servers, thus reducing latency. That's about as much detail as you need to get the idea.

p.s. I want to make steady consistent profits so I avoid anything crypto like the plague.

Edited May 14May 14 by IsaanT

4 hours ago, gargamon said:

I got my computer science degree in 1983. There was no AI, no concept of AI, no AI algorithms, etc.

AI dates back to Turing during the 1950s.

3 hours ago, Thingamabob said:

There is a lot of AI generated bull<deleted> on the internet.

There is also a lot of human generated bull<deleted> on the internet. So what is your point?

4 hours ago, gargamon said:

Rubbish. 50 years ago was 1976. I got my computer science degree in 1983. There was no AI, no concept of AI, no AI algorithms, etc

I agree that there was no AI as it is known today, however, the concept and early theories regarding AI were developed during the 1960s. Then the 1970s were pivotal eras for developments in AI, building on early theories to create practical "Expert Systems" and the foundations of modern neural networks. Some of the key breakthroughs that occurred during the 1970s:

• Expert Systems: Researchers developed systems like MYCIN to simulate human decision-making in specialized fields like medical diagnosis (Radanliev, 2024).

• Prolog & LISP: These logic-based programming languages became the standard for AI development.

• Backpropagation (1974): Paul Werbos introduced the backpropagation algorithm, which would eventually allow multi-layer neural networks to learn efficiently.

4 hours ago, Dexxter said:

There is also a lot of human generated bull<deleted> on the internet. So what is your point?

My point is that, in addition to the human generated BS on the internet, there is now also a lot of AI generated BS on the internet.

Humans have no future.

8 minutes ago, Jingthing said:

Humans have no future.

You post some weird stuff at times🤔

27 minutes ago, couchpotato said:

You post some weird stuff at times🤔

More like you're not paying attention.

We've created something with the power to be our masters.

9 hours ago, Thingamabob said:

There is a lot of AI generated bull<deleted> on the internet.

And on here too.

14 minutes ago, Jingthing said:

We've created something with the power to be our masters.

Every AI ,robot and data center needs these in prominent and accessible places.

8 hours ago, Rockyroad said:

Ai is only wrong due to humans putting wrong info online!

One of the AI companies is using Reddit as a source for training data. I noticed that some of the information is sourced from Reddit, and I'm not a fan of Reddit. Therefore, you have to be careful about the replies from AI. If you look at the bigger picture, AI can be dangerously biased if someone plants false data on the internet.

On 5/11/2026 at 10:11 AM, gargamon said:

Question 1.

How do you weed out the incorrect info from reddit etc? Or is that why sometimes answers are incorrect?

Claude responded: That's a great question, and yes — that's essentially one of the core reasons I can sometimes be wrong.

That's a great question, and yes — that's essentially one of the core reasons I can sometimes be wrong.

The short answer is I don't "weed out" bad information in real time. During training, Anthropic uses various techniques to improve accuracy — things like human feedback rating my responses, cross-referencing multiple sources, and filtering obvious low-quality content. But it's imperfect because:

• Reddit/forums contain both great advice and completely wrong advice — and it's not always easy to distinguish at scale

---

40 minutes ago, Effective altruism said:

One of the AI companies is using Reddit as a source for training data. I noticed that some of the information is sourced from Reddit, and I'm not a fan of Reddit. Therefore, you have to be careful about the replies from AI. If you look at the bigger picture, AI can be dangerously biased if someone plants false data on the internet.

I asked claude that very question and it was in the OP.

47 minutes ago, Effective altruism said:

AI can be dangerously biased if someone plants false data on the internet.

indeed

https://www.unz.com/cjohnstone/netanyahu-stresses-the-need-for-more-propaganda-as-israels-hasbara-budget-soars/

On 5/10/2026 at 10:10 PM, gargamon said:

In your dreams. Science is being killed at a government level.

In the U.S.

8 hours ago, Effective altruism said:

One of the AI companies is using Reddit as a source for training data. I noticed that some of the information is sourced from Reddit, and I'm not a fan of Reddit. Therefore, you have to be careful about the replies from AI. If you look at the bigger picture, AI can be dangerously biased if someone plants false data on the internet.

Like this forum. So much bad info.

9 hours ago, Effective altruism said:

One of the AI companies is using Reddit as a source for training data. I noticed that some of the information is sourced from Reddit, and I'm not a fan of Reddit. Therefore, you have to be careful about the replies from AI. If you look at the bigger picture, AI can be dangerously biased if someone plants false data on the internet.

I've not noticed any bias with Gemini except may mention American products as default

11 hours ago, gargamon said:

I asked claude that very question and it was in the OP.

I don't read your posts.

2 hours ago, scubascuba3 said:

I've not noticed any bias with Gemini except may mention American products as default

I tested a few Ais and gemini had the best info. Still it misses stuff.

2 hours ago, Effective altruism said:

I don't read your posts.

You're missing out.

On 5/14/2026 at 8:37 AM, gargamon said:

Rubbish. 50 years ago was 1976. I got my computer science degree in 1983. There was no AI, no concept of AI, no AI algorithms, etc. My earlier courses were done on punch cards. Terminal access to mainframes was very limited.

Agree with the AI part being nonexistent. I started working with mainframes in 1980 and did courses in IBM classrooms / reading manuals.

6 hours ago, FritsSikkink said:

Agree with the AI part being nonexistent. I started working with mainframes in 1980 and did courses in IBM classrooms / reading manuals.

In the late 80's the project I was working on was the largest software project in the world. We were well over 30 million lines of code when I left. The ”product” we were selling started at $10 million per system and delivery date was 18 months out as we couldn't build them fast enough. We were telling IBM what the specs for their next mainframes needed to be.

After I departed them the Americans took over the company and ran it into bankruptcy quite quickly. Sound familiar?

It's worse than I expected. There's no future to this stuff.

---

You said: so as a retired software guy I have a question about AI. Evryy time I input a new entry in a chat do you need to "recalculate" everything again?

Claude responded: Yes, essentially.

Yes, essentially. Each time you send a message the entire conversation history is sent to the model as context and processed again from scratch. There's no persistent memory of previous tokens — the model reads the whole conversation each time to generate the next response.

This is why very long conversations can get slower and also why there are context window limits — there's a maximum amount of text the model can process in one shot.

It's quite different from how a human conversation works where you just remember what was said. Each response here is essentially a fresh inference pass over the entire conversation history.

The upside is consistency — the model always has full context of everything said. The downside is it's computationally expensive and has hard limits on conversation length.

You said: well that explains the CPU/memory issues with AI

well that exxplains the CPU/memory issues with AI

Claude responded: Exactly — and it scales badly.

Exactly — and it scales badly. Double the conversation length, roughly double the compute. That's why:

Long conversations get slower

Running large models locally needs serious hardware

Cloud AI is expensive at scale — you're paying for a lot of matrix multiplications per token

Context windows are a hard engineering constraint, not an arbitrary limit

The transformer architecture that most modern LLMs use has attention mechanisms that scale quadratically with context length — so a conversation twice as long takes roughly four times the compute just for the attention calculations. It's an active area of research to make this more efficient.

For a retired software guy the mental model is something like: imagine having to reread every email in a thread from the beginning every time you want to reply, and the cost of reading scales with the square of the thread length. That's roughly the situation.