#SystemCard - Twitter Hashtag

over 1 year ago

#OpenAI quietly updated their #systemcard for #GPT45. Here are the main differences (thanks to o3-mini-high): https://t.co/TxKUszbMTf the Hallucination Evaluations (Table 4), the GPT‑4o numbers were updated (accuracy improved from 0.28 to 0.50; hallucination rate improved from 0.52 to 0.30). https://t.co/E4Jtqbgfhe the Preparedness Framework section, the original claim regarding “improving on GPT‑4’s computational efficiency by more than 10×” has been removed. The updated version now emphasizes that GPT‑4.5 is “our most capable GPT‑series release” while noting it does not add net‑new capabilities on most evaluations. https://t.co/yghQ4tdCcx the authorship and acknowledgments, several contributor lists have been modified: – “Daniel Levy” now appears among the Foundational Contributors. – New names (e.g. “Ananya Kumar” and “Charlotte Cole”) are added to the Research Core Contributors. Comparison screenshots below

1

0

90

Shaun Ralston

@shaunralston

almost 2 years ago

Yes, and that's such a good point, @KeyTryer! I think you'll find the full documentation in the GPT-4o #SystemCard interesting (with other audio examples, too). You're DEF right about this being a balancing act between reasonable safety and a full lobotomy. https://t.co/IOKKvcJWib

0

2

0

60

Shaun Ralston

@shaunralston

almost 2 years ago

Interesting; audio samples, description, evaluation and risk mitigation in the GPT-4o #SystemCard #Safety https://t.co/IOKKvcJWib

shaunralston's tweet photo. Interesting; audio samples, description, evaluation and risk mitigation in the GPT-4o #SystemCard #Safety
https://t.co/IOKKvcJWib

0

8

0

4

2K

NeowinFeed

@NeowinFeed

almost 2 years ago

By @pradeepviswav - OpenAI has released the system card and preparedness framework scorecard for its newest multimodal model, GPT-4o. It has implemented safeguards at both the model and system levels. #OpenAI #GPT-4o #SystemCard https://t.co/GAULDJjK6g

1

0

665

LLM Master @llmmaster

almost 2 years ago

¡#OpenAI ha lanzado #SystemCard! Ahora podrás evaluar la seguridad de sus modelos de IA de un vistazo. ¿Qué te parece? https://t.co/kxNxhiZsYy

0

3

0

147

Shaun Ralston

@shaunralston

almost 2 years ago

Check out today's @OpenAI report outlining safety work carried out before releasing GPT-4o #SystemCard https://t.co/hPxoIXBJeV

0

53

4

5K

Terri Nakamura

@terrinakamura

over 4 years ago

Lots of great information here — Is Meta's System Card the next step to safer algorithms? https://t.co/UoIFBh4HWa By Avi Gopani for Analytics India Magazine #instagram #meta #systemcard #algorithms #rankings