Top
Best
New

Posted by albelfio 17 hours ago

Introducing System One Models and Jev(typesafe.ai)
1476 points | 424 commentspage 6
initsecret 16 hours ago|
> [others] Output tokens: ~5x more expensive than input tokens.

> [them] Output tokens: FREE (too cheap to meter).

I'm very confused by this.

someguynamedq 1 hour ago||
This is not confusing
varenc 11 hours ago|||
The output tokens are just responses to your inputed questions and their probability. So relatively few output tokens. No unstructured text back in the response.
quotemstr 16 hours ago|||
They're not doing autoregression, so all the outputs are computed in one big forward pass. Very cheap.
ambicapter 16 hours ago||
I think OP is confused about "others" vs "them".
initsecret 16 hours ago||
they’re talking about two totally different things, right?
CompleteSkeptic 16 hours ago||
it's our output tokens that are free (under the system one / jev column)
niutech 3 hours ago||
How does Jev compare with encoder language models like BERT/RoBERTa/DistilBERT, which could also be used for text classification?
tylermarques 15 hours ago||
We had early access and found it to be pretty useful. Having a second form of verification, where you can ask multiple questions (in the form of Nouls) raised our confidence in the outputs of other models. [0] IMHO This type of model works incredibly well in concert with LLMs, not as a replacement.

[0] https://goodstartlabs.com/research/verification-is-the-bottl...

kevmo314 5 hours ago|
Don't your numbers suggest DeepSeek V4.1 Flash, for $100 more, gets you to slightly better agreement?
mentalgear 5 hours ago||
Overall this seems like a classifier that gives weighted scores per custom labels. It's certainly useful, but whether it brings higher quality than an LLM in structured output mode has to be seen in objective benchmarks.
StevenWaterman 5 hours ago|
Zero shot classifier indeed. Reminiscent of asking an llm a yes/no question, constraining the output to either yes or no, and looking at the logits directly

And each question is a separate single token model completion done in parallel

xynelius 13 hours ago||
The Doom demo looks impressive but was it a fine-tuned model? It's the difference between a cool demo and revolutionary tech.
copperx 13 hours ago|
Shouldn't self-driving be a piece of cake if it works this well for Doom? Or what am I missing?
pantelisk 12 hours ago|||
I think the doom demo uses a text representation of the world and it's basically, "projectile coming your way" -> "Strafe". "Enemy ahead" -> "shoot. So it works well when spawned in a room of enemies (as we see in the video).

If self driving is red means stop, green means go, and stay in your lane - then it would work great, but having to actually think and test which maneuver is optimal for a given situation while weighting safety, road rules, random unexpected actions and getting to your destination, I think it's a much bigger problem. A bigger model specifically trained on that maybe would do great, but then the output is not the constraint anymore.

But I haven't tried the model, so I 'm just ballparking and could be very wrong.

hamishwhc 11 hours ago||||
The model doesn't have image input capabilities (yet, it seems from the post), so for the Doom demo, a harness is extracting a bunch of structured information from the game (map layout, enemy locations, player ammo, health, etc) and providing it as a massive JSON blob to the model so it can make its decisions. This model _could_ be hooked up to make the decisions for a self-driving car, but it would need to be fed a structured blob of the situation around it, so all the computer vision problems of self-driving are still there. And that's before you get into the confidence and accuracy of this model.
jamilton 11 hours ago|||
Driving is more complicated than Doom, and it doesn't look that great at Doom to me.
paraschopra 6 hours ago||
I'm trying to understand what difference does this make over LLMs.

LLMs are universal simulators, their latents model the world. So I bet if you compare their logprobs with probabilities output by this model, it will be highly correlated.

Someone should do this quick experiment. I bet there won't be enough of a meaningful difference.

warpspin 15 hours ago||
Haven't seen any docs or so. Is this actually a general model, or does it need training on the the data set it answers? Finding it suspicious you never see some kind of prompt.

Edit: never mind, found https://docs.typesafe.ai/introduction/quickstart by now

CompleteSkeptic 14 hours ago||
1. yes a general model 2. no training at all 3. but it is focused on "System 1" tasks (more human judgment, less math reasoning)
zenlikethat 14 hours ago||
It's very generalized. Can't wait until everyone can see it.
aryehof 7 hours ago||
As a zero-shot classifier, I expect that effectiveness is dependent on the data trained upon.

Jev input … > Unstructured data (e.g. text) with an emphasis on structured program state.

What pre-training data/model is Jev based on? Surely result effectiveness is dependent (outside of one’s own input as “state”) on that?

torginus 15 hours ago|
I was thinking about something similar (maybe) - generally speaking, embeddings for LLMs tend to learn real world concepts - things like 'fruit' or 'France' or 'city' as directions in embeddings.

But in things like programming, most concepts are abstract - 'if hungry eat an apple' in programming terms would look like

'if hunger > 50 {apples--; hunger-=30;}'

and compilers work with 'concept erasure' - to them, tokens (which are like llm tokens) look like

'if var1 > 50 {var2--;var1-=30}'.

They don't care about how these things map to real concepts. So all the embedding directions used to encode real-world concepts are just noise to LLMs when programming. This greatly reduces dimensionality and training costs. So does a token representation tuned for programming constructs, rather than natural language would probably have a more efficient encoding.

ta988 15 hours ago|
Current models go beyond the simple embedding because you start to encode groups of concepts in the context-aware part of the model (attention heads or any other method). So it is never simply words/tokens in isolation anymore.
More comments...