• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

The AI discussion thread

Page 47 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
As long as they can’t solve the hallucination problem — which seems to be getting worse — how can LLMs ever be more than just tools of limited use? How do you justify the ridiculous sums of money, resources, and energy usage if this isn’t ultimately a solvable problem?

 
Racan said:
As long as they can’t solve the hallucination problem — which seems to be getting worse — how can LLMs ever be more than just tools of limited use? How do you justify the ridiculous sums of money, resources, and energy usage if this isn’t ultimately a solvable problem?

Click to expand...
I think having an isolated offline AI that's actually permitted to learn and grow would work much better than our current common LLMs available. The ability to change weights, correct hallucinations, infer preference, etc would help turn your shittiest free employee into a competent one.
 
My employer has banned the use of AI generally for anything. People were typing in too many keywords and technologies that interfere with our ability to keep our proprietary processes private. Moreons.
 
Google Gemini AI Glasses, aka Google Glass 2.0:

1. Looks normal
2. Runs Android XR
3. Comes our next year

Neat features:

1. Real-time language translation with subtitles
2. Memory assistance tracks where you left stuff, similar to Apple Tags & Tile trackers, but using the camera
3. Memory assistance also remembers your day, recapping what you talked about (ex. what someone told you hours ago) & what you did, including things like appointment reminders, so you basically have a photographic memory

No more Borg face:

1747499406752.png

This actually looks REALLY useful:

 
Miro AI node-based notes:

miro.com

Miro | The Innovation Workspace

Miro is the innovation workspace where teams manage projects, design products, and build the future together. Join 90M+ users from around the world.
miro.com miro.com

I use Plectica, but Miro has AI assist & more features, very cool!

 
nOOky said:
My employer has banned the use of AI generally for anything. People were typing in too many keywords and technologies that interfere with our ability to keep our proprietary processes private. Moreons.
Click to expand...

Yeah that is definitely a potential issue for non self hosted AI. If serious companies are going to use AI they will want to self host it. My company unveiled an AI but I played with it, and think it's just an API call to Gemini so nothing really spectacular and not usable for company specific use. I tried to ask it a generic question about a piece of software we use just to see if it knows anything about it and it did not.

Right now there are options for self hosted but not sure if they allow you to actually train it, I think they're static. Going to be neat when there are ways to train it, so it can be trained on company specific data. From a telco standpoint I could see so many uses for that such as getting info on buildings, previous incidents, who to call for a certain area, maybe even customer specific info etc. It could even learn alarm trends and such.

Of course this would need to be setup in a way where the AI recognizes access levels, so only those who have the right access would get the info.
 
VEO 2 showreel. For reference:

1. "The first Toy Story film (1995) required 800,000 machine hours and 114,240 frames of animation in total, divided between 1,561 shots that totaled over 77 minutes. Pixar was able to render less than 30 seconds of the film per day."

2. In Toy Story 4, the most expensive frame took 325 hours to render on a 4-core device.

3. Now you just pay a supercomputer & tell it what you want like magic lol

 
You must log in or register to reply here.

TRENDING THREADS

Back
Top