Create accountLogin

Recent searches

No recent searches

Search options

Only available when logged in.
eupolicy.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
This Mastodon server is a friendly and respectful discussion space for people working in areas related to EU policy. When you request to create an account, please tell us something about you.

Administered by:

Server stats:

216
active users

eupolicy.social: AboutProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.3.6

#llama

5 posts4 participants0 posts today
Public
BGDon

Good read: Beware of AI model collapse!

In an AI model collapse, AI systems, which are trained on their own outputs, gradually lose accuracy, diversity, and reliability. This occurs because errors compound across successive model generations, leading to distorted data distributions and "irreversible defects" in performance. The final result? A Nature 2024 paper stated, "The model becomes poisoned with its own projection of reality." nature.com/articles/s41586-024

"In simpler terms, when AI is trained on its own outputs, the results can drift further away from reality." as stated by Aquant.

Full article here: theregister.com/2025/05/27/opi #AI #AIErrors #AI_Model_Collapse #LLMs #ChatGPT #Llama #Claude #TheRegister

AI Neural Net
Replied in thread
Public *
Autonomie und Solidarität

"Nur noch heute können Nutzer der Plattformen von Meta, #Instagram und #Facebook der Nutzung ihrer Daten für die Meta-eigene KI widersprechen.

Meta will künftig alle öffentlichen Daten, also sowohl Posts als auch Kommentare seiner Nutzer, dafür nutzen, um das eigene "Large Language Model" mit dem Namen #Llama und die seit kurzem bei #WhatsApp fest installierte Meta-AI zu trainieren.

Zwar kann auch noch nach heute der Nutzung widersprochen werden, allerdings können einmal eingespeiste Daten rein technisch aus dem Modell nicht mehr entfernt werden.

#Meta will die eigene #KI in Zukunft für alles Mögliche verwenden.

Neben #Chatbots und personalisierter KI-Werbung könnten die #Daten auch militärisch genutzt werden. Eine Klausel, die dies verbietet, hatte Meta vor kurzem gestrichen."

Eine Anleitung zum Widerspruch findet ihr bei @netzpolitik_feed

netzpolitik.org/2025/meta-ki-j

netzpolitik.org · Meta-KI: Jetzt widersprechen oder für immer schweigenMark Zuckerberg will wieder mal an unsere Daten - dieses Mal um seine KI zu trainieren. Bis zum 26. Mai ist Widerspruch möglich: Wir erklären, warum das eine gute Idee ist und zeigen, wie es funktioniert.
#Datenschutz#Überwachungskapitalismus#AI
Public
Miguel Afonso Caetano

"Llama 2 is perhaps best known for being part of another scandal. In November, Chinese researchers used Llama 2 as the foundation for an AI model used by the Chinese military, Reuters reported. Responding to the backlash, Meta told Reuters that the researchers' reliance on a “single" and "outdated" was "unauthorized," then promptly reversed policies banning military uses and opened up its AI models for US national security applications, TechCrunch reported.

"We are pleased to confirm that we’re making Llama available to US government agencies, including those that are working on defense and national security applications, and private sector partners supporting their work," a Meta blog said. "We’re partnering with companies including Accenture, Amazon Web Services, Anduril, Booz Allen, Databricks, Deloitte, IBM, Leidos, Lockheed Martin, Microsoft, Oracle, Palantir, Scale AI, and Snowflake to bring Llama to government agencies."

Because Meta's models are open-source, they "can easily be used by the government to support Musk’s goals without the company’s explicit consent," Wired suggested.

It's hard to track where Meta's models may have been deployed in government so far, and it's unclear why DOGE relied on Llama 2 when Meta has made advancements with Llama 3 and 4.

Not much is known about DOGE's use of Llama 2. Wired's review of records showed that DOGE deployed the model locally, "meaning it’s unlikely to have sent data over the Internet," which was a privacy concern that many government workers expressed.

In an April letter sent to Russell Vought, director of the Office of Management and Budget, more than 40 lawmakers demanded a probe into DOGE's AI use, which, they warned—alongside "serious security risks"—could "have the potential to undermine successful and appropriate AI adoption.""

arstechnica.com/tech-policy/20

Ars Technica · Musk’s DOGE used Meta’s Llama 2—not Grok—for gov’t slashing, report saysBy Ashley Belanger
#USA#Trump#DOGE
ExploreLive feeds
About