• Security incident: ISF was recently accessed by intruders. Please change your password, and change it anywhere else you used it. Read more

Merged Artificial Intelligence

HansMustermann said:
I think the even bigger problem is not just being expensive, it's that you don't know how expensive it will be. You don't know in advance how many tokens a certain task will need to complete, and even less so after the several tries and tweaks to the prompt. You only know how much you've used when the invoice comes.
Click to expand...
Yeah. My employer is in the process of moving our entire data warehouse onto cloud-based Snowflake, and we will need tokens in order to access our own damned data that is needed to run the business. The token cost varies by the time the code runs, the volume of data returned, and the traffic load at that time. There’s no way to know what the cost will be, and building new code is going to be a nightmare because it’s never perfect the first time.

So far as I can tell, we’ve given all of our data to a gunman who is holding it hostage.
 
dirtywick said:
additionally, companies giving ai access to proprietary data and processes seems really foolish.
Click to expand...
Do you mean training AI models on proprietary information?

Or do you mean asking a trained AI model to evaluate proprietary information?

I don't see any particular problem with the latter. Contracting with secure third parties to handle some part of your proprietary business operations is already a well-established practice across many industries and services.

All our code is in GitHub. All our CI/CD is in CircleCI in the cloud. All our object storage is in Artifactory in the cloud. All our infrastructure is in AWS and Azure. All our data is in Databricks in the cloud. If I ask the AI model included in our DBX account to do some data mining across multiple disparate data sources, I'm not incurring any greater risk than I've already incurred.
 
Read and weep

Kevin Beaumont (@[email protected])

In an era where companies need to become more efficient and diverse they’ve basically picked the least efficient way to do it, with the biggest risks and highest costs - because everybody else is doing it. I know somebody at one of the big 4 who has written something in Claude that prompts...
cyberplace.social cyberplace.social
 
Kinda reminds me of a point someone else made on YT, so I can't claim credit. When the iPhone first came out, a lot of companies had a problem with bringing your own device to the office, especially one with a camera. Some even flat out forbade using a smartphone at work. But people liked them, they were useful, so people used them anyway. Meanwhile with AI you have to coerce people into using it, including, yes, by using the number of tokens in their job evaluation. What does that tell us?
 
I have not been following the AI threads...how much credibility is Roman Yompalskiy given? I think his actual quote was 99.9% doom within 100 years, but still...this reminds me of the Ai warnings from Dune, or the movie Forbidden Planet...

We're Doomed
 
arthwollipot said:
www.miragenews.com

AI Tackles Cocktail Party Problem in Hearing Aids

How do we recognize what one person is saying when others are speaking at the same time? This simple, yet peculiar question forms the basis of the
www.miragenews.com www.miragenews.com
Click to expand...
Googled Mirage News and other than loads of links to the site I didnt find much.

I did find an everybody wiki page for the site which states that it regurgitates "lightly edited press releases (a practice called churnalism)".


So until I hear something from a reputable source, I'm going to count this as barely edited advertising copy from an LLM company desparate to find a product to sell before they go bust.

PS the person cited seems to have been both studying in Brazil and the Netherlands for over two years according to his research gate profile.
 
Darat said:
I'd like a pair of earphones that could do that - I have always had great difficulty in hearing people in noisy environments.
Click to expand...
My Bernafon hearing aids have profiles like "Conversation in noise", "Conversation in loud noise" which are a little help. The "Conversation in car" is brilliant as it removes from of the fairly constant motorway surface noise.
 
Gulliver Foyle said:
Googled Mirage News and other than loads of links to the site I didnt find much.

I did find an everybody wiki page for the site which states that it regurgitates "lightly edited press releases (a practice called churnalism)".


So until I hear something from a reputable source, I'm going to count this as barely edited advertising copy from an LLM company desparate to find a product to sell before they go bust.

PS the person cited seems to have been both studying in Brazil and the Netherlands for over two years according to his research gate profile.
Click to expand...


They seem to have just taken the article from his engineering school's website but left out this line:

"Luan Fiorio recently defended his PhD thesis at the Department of Electrical Engineering."

www.tue.nl

Solving the cocktail party problem for hearing aids with machine learning

Luan Fiorio recently defended his PhD thesis at the Department of Electrical Engineering.
www.tue.nl www.tue.nl

So, apparently it is just a single doctoral candidate's independent research. The Mirage News site does actually link to the original article near the bottom of the page, though it might have been easy to miss.
 
Darat said:
I see that an open source model has already caught up with OpenAI's and Anthropic' s latest - Kimi 3 https://benchlm.ai/blog/posts/kimi-3-release-data-coming-soon

Oh dear....
Click to expand...
I don't see a problem here. It's still early days of LLM's. Right now we are at the stage of "it works, let's brute force it until we run out resources and then optimize". The next stage is innovation that dramatically reduces the resources required. The winners may not be those who started it or those who invested the most.
 
The problem is with the current "business" model not the technology* (I've been saying watch China for awhile, the international sanctions have been forcing them to be more efficient and to do more with less). At the moment there are two AI companies that everyone is relying on being able to pay their bills and commitments, and that is OpenAI and Anthropic. The problem they face from the likes of the open weight models is how do you sell something to make literally trillions of dollars of profit before 2030 when you have nothing unique to sell as any company that needs a foundation model can do it cheaper for themselves?

*Actually the technology is also the problem, ROI is still unmeasured as token charging is a mystery, and quality is still an unknown, we are all still being sold the future, this may be the next "fusion is 10 years away..."
 
A problem of genericized technology is the opposite of fusion. If it's already such a commodity that anyone can reimplement it for their own use, in the future it'll have even more impact without a monopoly to price gouge it out of reach of the little guy.

As for corpo promises, *shrug.* I don't think anyone here is arguing that the c-suite hype of earning a hojillion megabucks by q3 totes srs this time u guyz is in any way realistic. We're clearly in a bubble. Bubble's gotta pop, the earlier the better. Maybe the existing players survive. Maybe they don't.
 
Beelzebuddy said:
A problem of
genericized technology is the opposite of fusion.If it's already such a commodity that anyone can reimplement it for their own use, in the future it'll have even more impact without a monopoly to price gouge it out of reach of the little guy.

As for corpo promises, *shrug.* I don't think anyone here is arguing that the c-suite hype of earning a hojillion megabucks by q3 totes srs this time u guyz is in any way realistic. We're clearly in a bubble. Bubble's gotta pop, the earlier the better. Maybe the existing players survive. Maybe they don't.
Click to expand...

Yes and no - yes what they currently have is as you say the same as everyone else - my comment about fusion was more about the chase for AGI, it's always 2 years away.... What we do know is that AGI won't be achieved by buying more compute and running larger and larger models.
 
Beelzebuddy said:
A problem of genericized technology is the opposite of fusion. If it's already such a commodity that anyone can reimplement it for their own use, in the future it'll have even more impact without a monopoly to price gouge it out of reach of the little guy.

As for corpo promises, *shrug.* I don't think anyone here is arguing that the c-suite hype of earning a hojillion megabucks by q3 totes srs this time u guyz is in any way realistic. We're clearly in a bubble. Bubble's gotta pop, the earlier the better. Maybe the existing players survive. Maybe they don't.
Click to expand...
I don't think any of the LLM companies survive, even in the gentlest of pops. The question is does it take the likes of Google, Facebook, Amazon, Nvidia and Apple with them, given they've all drunk the kool-aid.
 
The problem with running local models is that they still require extra costs and are at best half as good. And I don't just mean the electricity. You're not going to get much done fast on a cheap I3 business laptop with integrated graphics. So either buy the most expensive 2 GPU computer for every employee, or essentially build your own servers. But most companies already prefer to outsource their IT to some cloud provider because it's a little cheaper. So you're back to using someone else's data centre, and at a per-use cost you don't control. You just changed it to being per compute power used instead of a mysterious token, but it's still outside of your control.
 
You must log in or register to reply here.

Share this page

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom