GPT store on Tuesday and GPT-4.5 right after? πŸ‘€

This account has already revealed some alpha information such as the announcement about the GPT store and the launch of Gemini:

gpt 4-5

Regardless of whether the prediction is correct or not, it seems we might get a 4.5 version before the big leap to version 5.

Reflecting on the evolution so far, the 3.5 model marked a notable advancement over GPT-3. In a similar vein, GPT-4 significantly outshone GPT-3.5.

What would you expect to see from a 4.5 model?

My expectations:

  • A marked improvement in handling inputs >4k tokens, which is currently quite precarious. [Reason to believe: Enhancing performance in this area isn’t overly challenging; the main hurdle is computational power.]
  • Notable enhancements in citation accuracy, reducing hallucinations of non-existent sources. [Reason to believe: Automating citation review is straightforward, but it demands extensive computational resources.]
  • Substantial advancements in tackling logical, mathematical, and general programming challenges. [Reason to believe: These domains are especially suited for progress through reinforcement learning, as they have definitive answers.]
  • The introduction of an additional multimodal feature, possibly involving video. [Reason to believe: Sam has repeatedly shown significant interest from OpenAI in this area.]
  • An end to relying solely on static past training data. This could mean including information updated with just a few weeks’ delay, fine-tuned weekly or monthly with current data. [Reason to believe: This highly sought-after feature was mentioned at OpenAI DevDay.]

Too optimistic?

@logankilpatrick Can you verify this?