this post was submitted on 05 Apr 2024
-27 points (22.4% liked)

Technology

59204 readers
3157 users here now

This is a most excellent place for technology news and articles.


Our Rules


  1. Follow the lemmy.world rules.
  2. Only tech related content.
  3. Be excellent to each another!
  4. Mod approved content bots can post up to 10 articles per day.
  5. Threads asking for personal tech support may be deleted.
  6. Politics threads may be removed.
  7. No memes allowed as posts, OK to post as comments.
  8. Only approved bots from the list below, to ask if your bot can be added please contact us.
  9. Check for duplicates before posting, duplicates may be removed

Approved Bots


founded 1 year ago
MODERATORS
top 8 comments
sorted by: hot top controversial new old
[–] CaptainSpaceman@lemmy.world 12 points 7 months ago (1 children)

Holup, this article is really saying a research paper can beat chatgpt?

[–] tsonfeir@lemm.ee 12 points 7 months ago (1 children)

This is why I think we should copy and paste the body of the article in its entirety into the post. Nine out of 10 times it’s some bullshit

[–] TheFriar@lemm.ee 7 points 7 months ago

OpenAI needs to watch out because Apple may finally be jumping on the AI bandwagon, and the news doesn’t bode well for ChatGPT. Apple is reportedly working on a large language model (LLM) referred to as ReALM, which stands for Reference Resolution As Language Modeling. Made to give Siri a boost and help it understand context, the model comes in four variants, and Apple claims that even its smallest model performs on a similar level to OpenAI’s ChatGPT.

This tantalizing bit of information comes from an Apple research paper, first shared by Windows Central, and it appears to be an early peek into what Apple has been cooking for a while now. ReALM is Apple’s own LLM that was reportedly made to enhance Siri’s capabilities; these improvements include a greater ability to understand context in a conversation.

The model is also said to be capable of processing onscreen content and more. That’d give it an edge over ChatGPT, which, although capable of processing image files and PDFs, can’t read your entire screen and react based on the contents.

Apple’s take on LLMs is said to be launching in four different sizes, referred to as ReALM-80M, ReALM-250M, ReALM-1B, and lastly, ReALM-3B. The letters at the end of each model name refer to millions and billions, respectively.

The researchers benchmarked ReALM and compared it to OpenAI’s GPT-3.5 and GPT-4. The worrying news for the Microsoft-backed OpenAI is that even Apple’s smallest ReALM-80M model is said to be all caught up to GPT-4, while all the ones above it are much, much better.

The paper remarks that ReALM shows large improvements over systems with similar functionalities, and the smallest model is up to 5% better when it comes to processing onscreen information.

While companies like Microsoft embraced the trend nearly as soon as it blew up, and, in Microsoft’s case, just made a fresh over $100B billion investment into the future of ChatGPT, Apple has kept quiet on the matter. Still, this isn’t the first time we’ve heard of Apple wanting to start getting into AI in a more substantial way — rumors late last year indicated that Apple had plans to give Siri much greater AI capabilities, and this paper is most likely the proof of those plans.

It’s unclear when Apple’s LLM will enter the mainstream market. However, it’s likely that Apple will discuss ReALM during WWDC in June.

[–] mokancan@infosec.pub 9 points 7 months ago (1 children)

Then why does their autocorrect suck so much?

[–] tfowinder@lemmy.ml 4 points 7 months ago

Because it's trying not to be Android

[–] Fiivemacs@lemmy.ca 5 points 7 months ago (1 children)

Try cooperating instead of 'defeating' everything. Fuck companies suck so much ass

[–] Squizzy@lemmy.world 3 points 7 months ago (1 children)
[–] Fiivemacs@lemmy.ca 4 points 7 months ago

Not when you are presented with the illusion of competition and their only goal is to maximize profits and minimize spending.