ygouzerh

Cam it be replaced by a Jev-like model? That might be better for latency sensitive tasks

I feel as well that it's the strongest reason. Even without AI, time pressure statistically already increase the error rate

I think here it's mostly that for normal business cases, we doesn't need to build one.

As a DevOps Engineer, I never once saw before the advantage of using a classifier. Now I see multiple parts of the stack where a better level of expressiveness will be useful (PR validations, Blue/Green validation, notification router for alerts, quick smoke tests, etc).

Nobody will give us the time and budget to build a custom classifier for these use cases, but a simple API call yes.

What are you using Haiku for?

Can it be that now they are getting optimized against benchmarks that are valuing logics, rather than human appreciation? (I am not an expert at all, just an idea)

Nice, thanks a lot, I didn't knew about this feature!

I think this is more a quote for engineers

One thing I am missing to be able to move out of Claude Code, is the auto mode (and the soft_deny and hard_deny settings that can be tuned), with it's classifier checking the output.

It's the killer feature from me personally, often when wanting to troubleshoot for example things like Kubernetes workloads. LLMs are now really good at it, but we doesn't want them to like delete a pod.

Other harnesses like Codex have often on static rules, like the allow/deny of claude code, that can filter out based on regex. It's quite good already, but sometimes the model can find a way to write something that wasn't anticipated, or in a convoluted way.

After, I guess it's something that can be added in an open-source harness like Pi, and add like this new Jev model or something else equivalent

Often politics, lack of vision, people being shuffled around, etc (e.g the stakeholder that decided of the acquisition just got moved to an another department, this kind of things)

That's a great point! It quite looks like the System 1 model of Physical Intelligence

Maybe a crash is what we need to reset the balance. We are spending a lot of money building shovels, but not enough to invest on the company that will use what will be mined.

If the money could flow back to the real economy instead of the AI economy, it might be healthier for the whole system.

It's crazy indeed. When working in company as DevSecOps, if we try to attempt pentest on prod without warning anyone, we will just get fired on the spot. AI Labs seems to have a free pass...

One thing that might be taken into account is the delay between ordering a plane, and the delivery (can take up to 10 years), so 1.5 years and 139$ million is worth investing into, to keep revenue flowing earlier.

Would a direct pitch to some HFT shops works better than kickstarter, to raise money for FPGA related work? They are ones of the main users in term of production usage. This project might be useful for them, for training / PoC / quick prototyping.

Quite interesting the response from MITRE: `It is not considered a security vulnerability because of how it requires a local attacker with privileges present to make it so`.

I didn't knew about that rationale. I thought Defense-in-Depth was all about that, preventing security incidents even when an attacker start to get some elevated privileges on the system?

If not, that's quite a practical reference for future CyberSec meetings.

For this one, I loved playing Dofus when I was younger. It was (hopefully still) a quite popular french turn-by-turn 2D MMORPG.

The main goal is PoE or PvP, but actually I ended up fully playing for the economy.

You can get a job with its own leveling system, like a farmer, cutting crops, collaborate with a baker who will make you a cut so that he can get baker xp, and then sell the bread in an auction based local marketplace for player to buy and heal.

You can as well just arbitrage: buy something cheaper in one town, and sell it in an another where it was more popular.

I ended up with a tone of paper notes, just calculating the margin of different product, e.g jewellery, so that I can buy the materials, calculating the risk of loss during the building process, partner with someone for some materials, and then finding where to sell the end product

It's the point that surprised me the most! We always used shallow clones, to speed the CI, I didn't knew that it got that much impact server side!

"Since AI excels at tasks like coding and some types of design, instructors can now assign projects that are much more ambitious." --> it's quite true.

For example, we never had time to learn well for example CI/CD in my years.

It's something that takes time to build (feedback loop is around 10 to 20 min for pipelines, instead of 1 min when coding something), and become quite costly for trial and error in term of time.

But it's now 60% of my work as DevOps Engineer, and a big part of other devs/lead devs work. They are a lot of good and bad practices.

We could imagine a class to learn about how to build proper CI/CD pipelines and experiment --> philosophies, different patterns, testing different caching strategy, etc

No worries, even on chromium, their webpage is just horrible. They seems to have focus only on mobile reading, on desktop its a mess.

You can use OpenRouter directly in Claude Code as well, it's quite nice!

For prompts that needs fact-checking, I like these days to use Perplexity directly instead these days. It's way faster than the default websearch tool + give a link to the reference directly.

You could hook it up with different tools, like real time information:

Oil + GPS + Web Search --> "LLM > You have 50 Km of autonomy. You can go today to this cheaper oil station, at 20 km, on your GPS road, instead of the one near your home. The one at 10Km is closed as well due to a local strike, I will avoid it too"

You can of course script all the scenarios + only use a TTS model. However, when plugging different systems together, I feel that it's the sweet spot where LLM is shining --> no need to pre-plan every scenarios that the user will ask, it can be done on the fly

I feel that it could be a nice addition for:

- People that didn't read the manual (actually almost all of us), like: explain a warning signal

- Or integrate different systems together: `I saw that on your GPS you want to go to this place, but in 2 hours it will be snowing heavily there. Please remember to bring your snow chains'

I felt as well that they miss their window. This kind of rectangular shape with keyboard + computer would be great actually for digital nomad / people going to coworking spaces, but looks way to heavy.

I don't think we need to wait 5 years, we can already see it know

Same, I am still thinking that it's a parody: going full loop back to programming

This resonates well with engineers: making something complex is easy, making them simpler is harder

This one looks shocking on the outside indeed! It's however a sales HR practice: lower salary and commission, but use the alluring treat that top performers will have a special exclusive trip at the end of the year. Many are crazy for it.

It's more a reward for a competition-style work mindset.

> It seems like AI has incidentally commoditized the complement (procedural knowledge), and in turn created greater demand for inputs like human judgement and "empathy".

I think as well there is currently the following: AI still needs an operator, in the same way than a bulldozer needs a bulldozer operator. And like with bulldozer, there are always something that business stakeholder wanna dig, now that digging is faster.

Crazy that it's a small lab becoming the frontier in term of moderation models, instead of Meta which is pouring dozens of billions into LLMs.

Meta would really benefit from work done on this front, however their model Llama Guards are quite lagging compared to the competition.