I believe there is a big opportunity for LLM guardrails due to the non- deterministic nature of the Transformer architecture.

However, the just announced Claude Cowork still warns humans to stay in control: https:// claude.com/blog/cowork-research-preview I assume this is because their non-human guardrails are not good enough yet to fully validate the output of an LLM.

What non-human guardrails does Axonflow employ to enforce a policy rule with X% confidence on a prompt / LLM output?

Thanks, this is a great question.

We intentionally avoid framing guardrails as “X percent confidence” checks on prompts or model output. In practice, probabilistic confidence at the text level has been the weakest place to enforce safety, especially once workflows become multi step and stateful.

AxonFlow’s non human guardrails are primarily deterministic and context grounded rather than model judgment based. Concretely, they focus on:

- authorization checks on actions, tools, and write paths rather than output quality

- permission evaluation per step using actual tool arguments and proposed side effects

- invariant checks on state transitions, for example whether an action is allowed given what the system has observed so far

- policy decisions that can halt execution entirely rather than degrade or retry

We do use probabilistic components in narrow, explicit places such as PII detection or risk classification, but those always feed into a deterministic policy decision. The system never proceeds because a model “seems confident enough.”

Human approval gates are not there because non human guardrails are insufficient in principle. They exist because some actions are intentionally irreversible or high blast radius, and no amount of model confidence should bypass explicit authorization.

So the distinction we draw is less about validating LLM output and more about deciding whether the system is allowed to move forward at all, given the concrete context and constraints at that moment.

The introduction seems to have AI sprinkled all over it: ..we embarked on a significant journey, ..in this monumental upgrade.

Archive link: https://archive.li/OmQIU

Beginners question but what is its purpose?

Diffraction grating (separates light by wavelength)

Personally I really enjoy doing analog photography. As the feedback loop is very long, technique is important. Also, since the number of pictures is limited, I always have to ask myself the question: is there actually a picture here?

Analog has all the tactile aspects I miss from DSLR and more, I definitely take more analog than I used to.

I also think there's a noticable gulf between fake analog and real, even on screen. The most skillful photoshop creations still looks totally off to me.

Phone + an old analog camera is an incredible setup imo

I usually consider programming books as a reference and don't "read" them. My way of getting value out of these books is remembering the high-level contents and diving deeper when there is the need (e.g, a coding problem that I need to take care of)

So sorry to hear about your losses. Sending you a virtual hug through cyberspace.

Because the syntax is relatively complex and it is difficult to judge which endpoints and definitions to use.

I learned SPARQL recently, and would agrre its complicated to get info out of Wikidata.

However, having read the article, they didnt have an easy time with scraping Wikipedia either.

So I'd probably still recommend people look into wikidata and SPARQL if they want to do this kind of thing.

Theres a few tools that generate queries for you, and some cli tools as well:

https://github.com/maxlath/wikibase-cli#readme

It makes Wikipedia better too, in a virtuous cycle, with some infoboxes like those that he scraped being converted to be automatically populated from wikidata.

Did we hug it to death? Got an error:

> Could not reach Cloud Firestore backend. Backend didn't respond within 10 seconds.

Great idea! Did you deploy a speech-to-text pipeline to achieve this? I always thought it would be relatively expensive to do podcast-to-text translation at scale (compared to the gains) but maybe I just didn't optimize it well enough :)

Not OP, but I've looked into AWS Transcribe [1] and at least their solution would begin to rack up quite a bit of a bill. From what I've seen, there isn't a great open source SST solution yet, although there do seem to be quite a few promising ones [2]. STT is one of the technologies I'm looking forward to most in the open source realm.

[1] https://aws.amazon.com/transcribe/pricing/?nc=sn&loc=3 [2] https://github.com/mozilla/DeepSpeech