Jev introduces a new shape of LLM(simonwillison.net) |
Jev introduces a new shape of LLM(simonwillison.net) |
<option>option A</option> <option>option B</option><endofoptions>userprompt<eos>
You can talk to an application and have it respond in real time with this combo. It’s clear this kind of general purpose intelligence may be a new development primitive.
However, at this stage, it’s difficult to work with for a few reasons. It’s API only, and you have to shape calling software to the way it communicates.
It’s not clear yet if what we’re missing is a new programming language, or some kind of harness or tool over the capabilities. LLMs were like this early on as well until better harnesses came along and reduced friction in their use.
Down the road, I highly suspect we’ll see:
- Intelligent context assembly using System 1 that summons memories as needed in LLM conversations and handles simple commands.
- System 1 programs that run in the datacenter and reach out to the request initiator on specific instructions like a CPU that has hit a memory barrier.And in my experiments even Qwen 3.8 has a hard time to consistenly conform to a schema, requiring retries, JSON cleanup etc, so to have a model of similar quality (SemIf et al) that simply cannot deviate from the schema by construction could be very helpful.
But I still need to experiment with either Jev/SemIf myself.
This is a valid point.
> And in my experiments even Qwen 3.8 has a hard time to consistenly conform to a schema, requiring retries, JSON cleanup etc, so to have a model of similar quality (SemIf et al) that simply cannot deviate from the schema by construction could be very helpful.
Literally every inference framework supports constrained encoding. You can make the model choose only from allowed tokens and you can infer only the first diverging token.
It's baffling to me that no inference provider actually exposes this functionality, so you have to run the model yourself to do it.
Although to be fair we don't know enough about the architecture.
Where I understand Jev to be a significant jump is that afaik the confidence scoring is actually derived from the normalised probabilities, and not a continuation in a chain of prediction masquerading as "confidence."
Broad questions like Is this resume good / score this city will ofcourse be biased but I think jev encourages more granular focused questions like Score this candidates Python experience / Rate this city for its food which then allows you to introduce your own biases in which questions you ask and how you combine their answers.
In this way I think jev like models can be easier to reason about for critical decisions.
Do people not feel like LLM speak (Claudisms) is infecting their own diction? Saying 'a new "shape" of LLM' sits so very poorly.
If so: no.
idk maybe this is a feature? if some LLM model is wonky, it won't generate right json format, so I'll know for sure
but for Jev, it's 100% right format, so I can't know if it's gone wonky or not