eigen-vector

Yegge is basically posting straight up slop these days. He's making some extremely confident claims about google after talking to one person. Didn't stop to think if that person is the most informed on this.

Somehow yegge also has problems with the adoption curve being consistent with other companies. Whatever that means.

Thank you for taking a look! Yeah there's definitely a few tells which are noticable if you look for it :)

Thank you for trying! I first built it as 'detect the human' response, but that was counter to the 'slop or not' framing. Yeah I'm also observing the same based on the first few hundred people's results. The harder models seem to write almost too well and that's generally not how humans write on the internet unless it is a blog post/essay. The easier models seem to be the ones that are tripping people the most.

Ha :) I'm not building models, nor am I affiliated with any big labs. The idea is to use this to educate people how to spot tells of AI writing. Although like any data that's made open this can be used to train future models as well I suppose.

thank you so much for taking a look :) Yeah you'd be surprised how difficult it can get to spot nuances sometimes. Sometimes, there isn't any nuance at all and the AI is just as good at writing about something pretending to know about the topic.

You'd be surprised at the nuances we tend to miss :)

This time around I prompted the models not necessarily to be adversarial - i didn't ask them to try and fool the reader. But i gave them contextual info - something to the effect of "you're a user posting on hacker news"

Thanks for checking it out! The color signal is useful feedback. Let me think about it and rework!

Yeah there are some very obvious tells, but the models that are most capable are very good at writing like human.

Especially when the human responses for reddit or HN prompts were presumably made after reading the content of the article or the post; whilw the model is simply going off of the title.

Thanks for checking it out! The obvious ones are (hopefully) weaker models :) but yes my experience has been unless you're engaging with human written content consistently the line really blurs easily.

Such a beautiful thing :)

absolutely. May be the best way to do this would be some kind of a recipe store where the user can run (we can fuzzy match?) tshark oneliners. I'd love your thoughts on what the easiest/quickest integration would be.

Thanks a lot! There is a bookmark and export feature that does what you're asking for!

Sorry that was a miss on my part. I fixed it in the latest main. Will recognize both in the next release.

I've added images to the readme :) Thanks for your feedback!

You are not wrong :)

Will do! I agree

This has been addressed and is in main now :) thanks again!

All this has been pushed to main. Thanks again!

So sorry I couldn't resist :) But hey this is a great way to get your kid interested in networking!

Thanks a lot. No reason to have upper case letters besides forcing intent and not move on accidental key presses. However, I have 'q' not mapped to upper case so I missed the obvious footgun :)

thank you :) I would be lying if I said the name wasn't the spark. The project lives to serve the name.

Thanks a lot for trying. My experience with packet inspection is similar and that's what resulted in me trying to build a simpler plain language UI companion.

• Back navigation: good call. I'll make "Esc back" more explicit everywhere and add Backspace as an alias for back (and mention it in h help).

• Enter on Domains not working: it should drill down to flows. If you can share your OS/terminal + whether you installed from release vs cargo install, I'll try to reproduce and fix in the next release.

• "Clicking" wording: agreed — mouse isn't supported right now. I'll change Ul copy to say Select (and keep "Enter = drill down").

• Packets screen truncation: yep, needs scrolling/paging. On the list views I already do r/; l'll add page scroll and horizontal handling where it makes sense.

• Numbered weird options: great idea - I'll map 1..9 to jump-select and Enter (or open directly).

• Domain names in packets: yes, I want that. I already collect DNS/SNI/HTTP host hints; next step is showing hostname labels alongside IPs when I have more confident mapping.

• "f stream" definition: I'll clarify it as "Follow stream (reassembled payload)" and add a glossary/help entry.

• Inspecting encrypted/compressed data: totally fair. The tool can't decrypt TLS without keys, but it can make it clearer what's happening (SNI/ ALPN, sizes, timing, resets/retransmits). I'Il improve "Explain" to say what's possible vs not.

Super useful feedback thanks!

Yeah, right now it's closer to "triage for non-experts" than "full teaching tool," and l agree there's an awkward middle where it assumes you recognize some concepts (flows/ports/latency) while trying to help with the drilldown.

The direction I want to push it in is exactly what you're describing; make it a learning tool, where each detector/view answers: 1) What am I seeing? (plain language) 2) Why might it matter? 3) What's the next click? 4) What term should I learn? (glossary link)

If you're about to teach a Wireshark module next week, two super useful things would be: • 3-5 common lab prompts you give students (e.g. "identify the DNS failure," "find the top talker," "spot a TCP reset," "why is this slow?") • one small pcap you already use (or even just describe its scenario)

I can tune Overview/Weird/Explain around those and make the guided layer feel like the main product rather than a thin overlay. Also: if your students are GUl-only early on, that's a good callout - I should improve the README to frame Babyshark as "guided analysis," not "terminal is easier than GUI."

I'm also happy for your students to get hands on by sending PRs for things they wish are intuitive from the get go.

This isn't meant as a replacement for tshark. It actually uses tshark for the live capture part.

tshark is the engine; Babyshark is the guided Ul on top of it. • tshark: raw packet/field dump + powerful filters, but you have to know what fields to ask for and how to stitch the story together. • Babyshark: gives you an opinionated workflow (Overview → Domains/Weird → Flows → Packets/ Stream) with "explain/why it matters" text, curated detectors, and one-key drilldowns.

For live capture, Babyshark uses tshark -T fields to extract things like DNS qname / TLS SNI / HTTP host; for offline PCAP it parses enough to build flows + summaries.

So: if you already live in tshark one-liners, tshark is faster. If you're trying to understand what's happening or teach/debug quickly, Babyshark is a nicer front-end.

I'm not trying to say it's better than the GUI but it hopes to be more guided. it’s *opinionated* about the first 60 seconds:

- *Overview dashboard*: immediately surfaces top talkers/flows + “what should I click next” instead of dropping you into the full packet list. - *Domains-first pivot*: `D` shows hostnames and lets you jump from a domain → the relevant flows. It also works when DNS answers aren’t visible (DoH/DoT/cached) by using observed IPs from SNI/Host flows. - *Weird stuff*: `W` is a curated set of “likely problems” (retransmits/out-of-order hints, resets, handshake issues, DNS failures when visible) with a short “why it matters” and a drill-down. - *Explain*: `?` gives plain-English hints for a selected flow + suggested next steps (follow stream, filter, pivot to domains/weird).

So it’s basically a guided triage layer on top of tshark/pcap data, with the “where do I start?” path baked in.

If you’ve got a specific teaching use-case (e.g. “why is this slow?” or “which host is generating traffic?”), I’d love to tune the Overview/Weird detectors around that. Open to PRs as well.

Thanks for the look. Babyshark is inspired by a bunch of terminal tools (termshark included), but the focus here is different: domains/weirdness-first drilldowns + "explain" + live-mode hostname hints (including observed IPs when DNS is encrypted/cached). If you try it and have specific gaps vs termshark, I'd love concrete feedback /issues.

Beta open now. The files are local and everything stays in your machine and limited context get sent to LLMs, you can read the security notes and threat model on the site. Commands API supports only anthropic for now. You can bring your own API key or use the demo key with limited usage. Plenty to get a feel for it.

A quick 30 second experiment to run:

1.Paste a small “Key Features” section

2.Type /// tabulate the features below into a recap table

3.Keep writing while it runs → output appears in preview → click “Insert into source” if you want it persisted

What Margin does today:

• /// commands that run asynchronously while you keep writing

• Generates structured artifacts (recap tables, Mermaid diagrams, charts)

• Clickable command status (running/done) + rerun/copy/insert/remove

• Export resolved markdown / copy rich HTML

I’d love feedback on: (1) does the “doc compiles into artifacts” mental model click? (2) what’s the first command you’d want as a writer?

Stack is react + tauri. Thanks!

Hey surprised to see this here. I made Drift. Thank you for sharing!

Just like how you can share Claude chat sessions there's a lot of value in sharing Claude code sessions as well. So I built claudereview.

You can privately or publicly share your Claude code sessions. With a handy link for PRs or blogs or socials.

Reviewers see how the code was built, not just the diff. With data encrypted at rest and private links which you can choose to password protect, and a clean and fast TUI inspired UI.

Features include:

• Key Moments: files created, modified, commands run

• Inline diffs for every edit

• Jump between tools (Edit 10/29 ← →)

• Full-text search

• Dark/light mode

You can install from npm/bun and it also has an MCP server you can use to share sessions inli ne. You can also add a /share slash command to do this while you're coding.

It's open source and free - https://github.com/vignesh07/claudereview

I'd love your feedback!

I built a tool to help people quickly make sense of their AWS/GCP/azure usage. And point out month to month differences, what drove changes, and what they can do to save money. You can also "chat with your usage report" and get recommendations.

It's in beta, so it's free with some usage limits that reset daily. Would appreciate feedback!

It costs nothing to be empathetic.