Uploaded September 2026 | Updated November 2026, 3 days ago
Everyone is hyping Jev as a brand new kind of AI, but I have seen this before.
▸ Here is what Jev does: you give it some text and a list of options, and it gives you back probabilities
▸ One pass, no text generation, no hallucinated answers
▸ That's exactly what encoder models like BERT have been doing for years: a classification head with the softmax
▸ And proof: Laya is the open source alternative. It is literally ModernBERT with a decision head, 421 million parameters, and it runs on your own GPU
▸ When ChatGPT came out in 2022, everyone said encoders are dead. They are not. They have just got a new name: System 1
So here is the lesson: if your LLM call ends with "pick one of these labels", you don't need an LLM. You need an encoder.
Full detail is available on kgptalkie.com
Follow for more AI and LLM breakdowns.
#Jev #Laya #ModernBERT #BERT #Encoder #LLM #AI #Shorts
Everyone is hyping Jev as a brand new kind of AI, but I have seen this before.
▸ Here is what Jev does: you give it some text and a list of options, and it gives you back probabilities
▸ One pass, no text generation, no hallucinated answers
▸ That's exactly what encoder models like BERT have been doing for years: a classification head with the softmax
▸ And proof: Laya is the open source alternative. It is literally ModernBERT with a decision head, 421 million parameters, and it runs on your own GPU
▸ When ChatGPT came out in 2022, everyone said encoders are dead. They are not. They have just got a new name: System 1
So here is the lesson: if your LLM call ends with "pick one of these labels", you don't need an LLM. You need an encoder.
Full detail is available on kgptalkie.com
Follow for more AI and LLM breakdowns.
#Jev #Laya #ModernBERT #BERT #Encoder #LLM #AI #Shorts










