Without having read the article, I would say it demonstrates that benchmarks alone, without knowledge of how a model was trained (c.q. constructed), cannot reliably determine whether or not a model has certain capabilities.
Yeah. Also, we think that in certain scenarios, it's possible for "cryptographic patterns" to arise in model weights via normal training, especially if models are trained against things such as honesty probes.
Such a click-baity name. Open challenge to anyone to use this framework and reverse engineer the internal APIs iPhone uses to talk to Siri or shortcuts or app Intents.
Nah sadly not. It's very slow, fails to respond half the time (just decides to disappear, ot "we ran into a problem"), it's nowhere near the full duplex voice of GPT Live - Siri is still some dumb text model behind a TTS model, and does not work like modern agents which can take chains of actions for me. It's pathetic.
PSA: Do not eat the yellow snow. There are not 15 hot singles in your area looking to meet up. Do not click mystery "Link to download" blobs from strangers.
For what in particular? You can read about Pegasus [0], although that page doesn't even remotely deal full justice to the extent of how widely it's deployed (e.g. basically all Russian opposition was/is tapped with it and somehow it doesn't even mention this). LEA of a lot of countries have some way to break into locked iPhones, such as Cellebrite's tools [1] or GrayKey [2]. I mean, even the popular US body cam channels on YouTube casually mention unlocking perpetrator's iPhone using the "forensics methods available to the police" like it's no big deal.
Operation Triangulation [3-8] had two separate attack chains, one of which is redundant, implying it's here just because it's a part of a bigger modular suite and the attacker can waste decades old 0-days just like that, having a ton more in reserve, and also something that looks awfully like an intentional hardware backdoor. What's fascinating about this attack is that it should have obliterated any vendor's reputation as being highly secure or government-resistant, yet we still have people thinking iOS is secure, and Apple had worldwide media coverup and multiple credible hardware experts unrelated to either Apple or ARM saying the backdoor is not intentional but more like a "genuinely forgotten debugging feature" (guess how a good backdoor should look like...), see the HN discussion [9] and . Despite Steve Jobs passing away long ago, the reality distortion field around Apple still seems as strong as ever.
Obviously this is just a tip of the iceberg, and how many of these are in the wild or how many backdoors the hardware still has is unknown. I'm sure that with better AI capabilities closed source vendors have many big surprises coming in 2027-2028. The availability of public jailbreaks is unrelated matter, they're necessary to support third-party stores against Apple's will but that's clearly not feasible. Samsung phones didn't have custom ROMs for the same reason, for example.
Wrong thing to index on. Always bet on the slope, not on points in the slope. Things will only get better and better and factor in that downward slope thought while taking actions and pick ones that are favorable. The folks who focus on a point in time will be left behind, period.
Why show an intelligent UI within a chat interface. That’s pathetically lame. Better to go away from chat interfaces to rich visual interfaces. Doesn’t sound intelligent at all to me
Pretty interesting but I wonder apart from sovereign use-cases in EU, who uses Mistral and for what? How good are they compared to frontier open models like Kimi?
I agree sovereign markets themselves can be huge but just curious.
Is there an open standard or something for people generating images to encode them or is the standard private and only shared with frontier model builders? That's a shame if latter. But pretty cool standardizing technology.
This is a false comparison. What Jev does is fundamentally different from what LLMs are doing as a judge.
For a quick primer, Jev is able to provide confidence scores on its classification, i.e. it is able to calibrate how well it is able to predict. Being able to predict in a distribution is different from being able to calibrate confidence of the predictions which should happen from the question or domain distribution from which the decisions are predicted - being able to do that is tough and is not same as using LLMs logit probabilities which are predictions in the vocab space. Though both are loosely correlated and might converge as LLMs keep getting better, the former is a much stronger decision-making signal than the latter. Jev not beating LLM-as-a-judge might be due to various other reasons such as world knowledge etc, but Jev as a concept will always provide more reliable decisions / outputs than LLM-as-a-judge giving a scalar score.
Claude and GPT tell me all the time what they are confident and what they're not confident in. This is partially useful, but it's not a guarantee of anything, since they can be confidently wrong or unsure and correct.
Same with Jav.
Arguing that Jav in some way more useful than current LLMs because it includes a confidence score smells of someone who hasn't updated their priors since GPT-4/4o and it's particular issues.
Jav is cool because it's very fast, very cheap, and "good enough". That's it.
The benchmarks show it is on par or worse than even older LLM models.
* With jev-sec-bench the confidence numbers reached roughly the same as Qwen2.5-7b-instruct (confidence was off by about 6% on average in both cases).
* On the OpenRouter Banking 77 benchmark Jev did the worst out of all the models, being about 25 points off from reality.
I also think the fact that Jev hasn't published any real papers or actual technical details should make people a bit more skeptical than they have been.
A low-latency calibrated classifier is very valuable. Speed is not at all a blocker. You could've changed the LM-head of an existing, say 3B parameter or even 30B parameter LLM or do speculation to easily get Jev like outputs - achieving speed is not at all a problem.
It is about getting calibrated outcomes. Jev is able to provide that, which is a major unlock
Partially so. But the fact that we're comparing Jev to SOA LLM and pre-trained classification means it's essentially satisfied its reason for existence.
The benefits of Jev, and the reason I'm excited about them:
* Pre-trained/tuned to provide only structured output
* Small, fast, essentially trivial to locally run - this alone makes them an actual candidate for real-world planning systems. Laya is a local-first 400 million parameter model, for example.
* Their confidence scores are useful, even if those scores are not correct/true probabilities.
For me and my work, they look like nearly turn-key, tiny, fast, locally-hostable models that can manage state transitions in a deeply autonomous system.
LLM-as judge requires, well, what we know as a full LLM. A fully trained classifier model requires fully training - something you cannot / won't do for a embedded autonomous planning system (by contradiction - if this worked, we'd have used it everywhere already!). Totally unfair comparison in my mind.
> With llama.cpp, you can give it an ebnf grammar and constrain the output to any formal grammar you wish.
With llama-server, you can use basically any GGUF model as a Jev-like classifier. See Pi.dev's codemode and the accompanying llama backend to the "classify" function. The advantages of Jev-like inference are that:
- You generate 1 token per question in parallel, instead of sequentially generating a bunch of JSON punctuation. So you get much better latency and utilization.
- You can use a custom "sampler" that sees the raw logprobs for these tokens, including all the tokens that might have been generated. So you can generate some "probability" or "confidence" numbers based on how likely the model was to have generated a different answer.
So really, no difference? It's just a nice API with some computational efficiencies.
> And how is the confidence score any different than the softmaxed token probabilities you get out of running every llm?
You're asking me how a structured output which assigns probabilities to classifications of the input text is different than the next-token weights which are calculated while generating that output?
One is internal (token prob), one is output generated by that internal mechanism e.g.,
You have a ebnf grammar constrained to evade, land, search etc. Quite literally what's the difference between the log probs between those options and whatever number a very tiny model is coming up for the confidence.
Copy. The mechanism is probably similar once we abandon the "LLM that is asked to generate json" path. But the grammar alone doesn't ensure the output is a probability vector does it?
Both use the softmax operation, but "probability of emitting a given label" is not the same as "probability the decided action is correct". OTOH, taking an action to sample something to come up with "probability of emitting a label" is in fact the whole shebang if the label is the next action. If it works it works.
Who cares, this distinction is small.
For me, a planning guy, either your suggestion or Jev-like would work just fine, and you better believe I'll be testing both as I'm quite excited about being able to use less-structured, untrained systems in decision pipelines. Thanks for the heads up.
I agree that there are differences between Jev and LLM-as-a-judge (e.g. Almeida's assertion that LLM probabilities have been irrevocably biased by RLHF), but I am not sure about your description of 'confidence'. Perhaps I misinterpret you or the docs, but I understand it as simply being computed from the probability distribution: https://docs.typesafe.ai/confidence#how-confidence-is-calcul...
I don't know about Jev, but in my experience, 90% of the time, the confidence score such machine outputs (e.g. simplest case a kalman filter) is bogus because the model is wrong. Or not even wrong, but just not perfect.
Mathematicians build this, and they love the beauty of it so much that they loose touch of reality.
Usably good calibration is hard. It's hard with logistic regression, it's even harder with linear support vector machines, and it makes me question my life choices with non-linear models.
It's not necessarily because the model is wrong. I've had trouble getting useful calibration out of models with an F1 of 0.9. The fundamental problem is twofold. First, it turns out that [0.0, 1.0] is a much larger set than {0, 1}. Second, the kinds of use cases where you care about calibration tend to be fussy and demanding.
As a fun anecdote, I got a chance to ask the person who invented the method scikit-learn uses for calibrated SVMs if he had any advice, and his answer was basically, "good luck."
All that said, sometimes you don't actually need calibration; you just need decent ranking. "Items with a score of 0.9 should be more likely to be positive than ones with a score of 0.5," is an easier requirement than "90% of items with a score of 0.9 should be positive." But I've had trouble getting that out of highly non-linear neural models, too, because they oftentimes produce results where the relationship between score and probability of being in the positive class is not even remotely monotonic.
But I haven't poked at Jev like this either, so I do have to allow that maybe they've found the secret sauce.
This is hocus-pocus, Jev is basically a dumber LLM constrained to certain problems, it's not magic.
Jev's "confidence" scores are just as valid as an LLMs confidence score, i.e. just as likely to be wrong or hallucinated as any LLM.
Ultimately, people are somehow thinking Jev is mysteriously more likely to be right when ultimately it's another non-deterministic magic box except this time with a stricter interface slapped on top of it.
LLM’s are not able to give confidence scores, they make them up.
Your AI driven app is making that up. Your product manager and executive team’s demand for confidence in the UI is a totally fictional cosmetic telling them nothing. Your company sold bullshit confidence to your clients.
I’ve done this for many organizations that “formed a new team to work with the CTO on their AI strategy”, and the trappings are the same
You can have an LLM tell you how much of a schema it was able to get information about. And derive a “confidence” or level of compliance from the completeness of the schema
But this is layers upon layers of cruft that a classification model wouldn’t need
Softmax over logits doesn’t give calibrated probabilities either as a rule. I don’t want to comment specifically on Jev but as a rule it’s very hard to get good calibration because it’s somewhat in tension with minimizing training loss for neural networks, e.g. Guo et al (2017) https://arxiv.org/abs/1706.04599
While I know there are ways to improve calibration, I’d personally want to see a lot of evidence the probabilities were actually more meaningful before trusting them. I agree of course that asking an LLM to provide a confidence estimate is meaningless.
Yes, but they are not able to act on their decisions, or ask for elaboration.
Maybe someone did order those enlargement pills, and an agentic model would be able to scan email history and determine that, when classifying an email as spam.
LLMs can provide confidence scores. Just ask it and it will provide.
Are those scores accurate? Probably not. But are Jev's scores accurate? They certainly aren't with a torture test like "I've rolled a fair die. What is the result?" (It comes back with ~80% confidence on 1 in my testing.)
It seems like people are very enthusiastic about Jev giving scores for its results, but I'm not seeing a whole lot of attention paid to how good those scores are.
That's an interesting test, but is it really what jev is supposed to be good at? I mean, this is a genuine question because I don't understand jev, but wouldn't it need a more constrained question?
Probably not. But it does seem to have some idea of probabilities if I get a little more specific. "I've rolled a fair die. I see several dots on the top." Gives me 46% for 6, and about 16% for 4 and 5. If I say "The top is mostly white." then it says 1 with 50% confidence... and 6 with 31%.
Interestingly, if I add a "unknown" option then my original question says unknown with 100% confidence. "I see several dots on top" gives 89% unknown. "The top is mostly white" gives 30% unknown and 37% 1.
So it definitely has some ability in this area, but it's iffy. This isn't exactly what it's made for but it doesn't seem too far off from the examples like taking in a customer email and deciding which team it belongs with.
LLMs had ability to do this from the start, but at some point providers stopped exposing this api to prevent distillation. Just write the prompt, pass the data and sample 1 token returning logits, OSS inference engines can do it easily.
all llms output confidence scores on their tokens. That's literally all they generate. Picking an actual next token is a (semi) random sampling of those confidence levels.
Yes apple intelligence is pretty bad. https://alma.inc/ aims to be what Apple intelligence should have been by building a new interface to your computer. And certainly, it won't take up much disk space.
