Areena_28

The moat was never really the code....it was always the understanding that produced it. All those failed experiments and design iterations that led to the algorithm... that knowledge lives in your head, not the repo. Someone can clone the implementation, but they can't clone the reasoning behind every decision you made.

The audience problem is separate from the IP problem and worth solving on its own, start writing about the problem you're solving before you open-source anything.

Yes, and the main change is i now weight specificity much more heavily.

The things that still feel trustworthy are very specific details, opinions that could get someone in trouble, and writing that has a clear point of view. Generic correctness is cheap now, so it doesn't carry the same weight it used to.

For computer vision and ML work specifically, RHEL doesn't give you much advantage over Ubuntu or even Fedora for personal use. Most ML tooling like PyTorch, CUDA drivers, Jupyter is better supported and easier to set up on Ubuntu, and the community resources are much larger.

if you want something RHEL-adjacent without the overhead, Rocky Linux or AlmaLinux give you the same base for free and are worth a look.

damm true haha

Judgment in ambiguous situations is the one thing that's held up consistently. AI is good at defined tasks, bad at knowing when the task definition itself is wrong.

Also, deep domain knowledge is the other one..... knowing what good output looks like in your field is something models can't fake convincingly at the edges.

Thank you, someone pointed out this. Using claude has become a pain in my ass now. I used to use it for writing but now it has lost its creative thinking. In addition, its free plan daily limit gets easily exhausted with 2 prompts...makes me think after getting all the hype, what happens to these platforms

I’ve noticed this with model upgrades too: sometimes they improve the “thinking” but lose the tightness of the writing.

Most of these tools aren’t defensible because of AI, they’re defensible because of workflow capture. The model is the easy part now, the hard part is owning the context, the data, and the place where work actually happens.

That’s why a lot of them feel same kind on the surface. Same models, slightly different UI. But the ones that stick quietly wedge themselves into a real job and make it 5x faster or cheaper.

Building Gordon (https://trygordon.ai/) because most companies “have security” but no idea what their actual risk is. We’re building a platform that helps you pass audits faster, lower cyber insurance premiums, and makes cyber less of a fire drill.

Its like one place to see risk, catch threats, test what breaks, track vendors, train people, and not get hacked at 2 AM.

My perspective is different here. You were right about the VR thing, but the AI tools on which we are relying today will no longer be useful for us 2 years later. I'll give it a term of "graveyard of AI startups". the gap is more about timing, every AI tool which is being made today is built to give the capability of 3-4 combined AI tools.

The graveyard is very large and it will keep expanding.

hahaha, you are so right!

The disruption is 101% happening but it's mostly invisible from the outside. iI think its not new apps replacing old ones rather the same apps doing things that used to require 3x the headcount.

That said there are a few genuine examples

Cursor basically replaced how a lot of developers write code, Eleven Labs made voiceover work that used to cost thousands into a $5 task, and Gamma turned presentations from a half-day job into 10 minutes none of these replaced entire industries overnight but they each quietly made a whole category of work much much cheaper.

Yeah, i noticed it too. Something feels off with the reasoning on complex multi-step tasks compared to a few months ago. hard to tell if it's actual regression or just expectations creeping up as you use it more.

Been mixing Opus with Sonnet depending on the task. Sonnet handles most things well enough and Opus for anything that genuinely needs deeper reasoning. Try it out, may be you find it useful

I'll speak on the co-founder question, try out the the trial period approach. Give someone a 60-day paid project with a clear deliverable before any equity conversation. You learn more about how they think and execute in 60 days than in 10 interviews.

Impressive! FSI courses are genuinely good but the raw format makes them harder to stick with than they should be.

Two things you can try: dispute it with your bank as an unauthorised hold if any of those payments were card transactions, and file a complaint with the DFSA if your business is in a free zone. Stripe does have UAE regulatory obligations and a formal complaint sometimes moves things that support tickets don't.

The way we handle it is keeping a small set of fixed test cases that we never change. Like same inputs, same expected outputs. so when we tweak a prompt we run it against those first. if it passes the fixed cases and feels better on the new ones, we keep it.

Never experienced BBS firsthand but from everything i've read about it, the thing that stands out most is how intentional the interactions were. you had to actually want to be there. no algorithmic feed pushing content at you. The people who showed up were genuinely curious about the same things. This is complete opposite of how online communities work now where half the participants didn't choose to be there, they just got recommended in. honestly, every social platform is giving us things which we dont need but the algorithm is making sure we needs it

Honestly the leverage you have is the parts they can't vibe code. I mean DevOps, architecture decisions, anything that breaks quietly in production. i'd focus there and let them find out the hard way what happens when nobody understands the system they're shipping.

The joy thing is real, though. once a client stops trusting your judgment, the relationship is already broken. sometimes the right move is to document everything clearly, hand it over properly, and walk away before it becomes your reputation on the line. I'd like to know more opinions on this

Been using Aegis for a while now, the encrypted backup alone makes it worth it over Google Authenticator. losing access to everything because you switched phones once is enough to make you take backups seriously. Ente is worth a look if you want cloud sync without giving up control of your keys. haven't tried Raivo personally.

Nearly missing a passport expiry is the kind of thing that happens to everyone at least once and nobody builds anything about it until it happens to them. good that you did something about it.

It surprised me when i got to know, i wonder how long this went undetected. Compliance certification exists so companies don't need to audit their vendors themselves, but here the whole certification is fake.

We've been building in the GRC space with Mitigata, and the pressure to make compliance "fast and painless" is real. And i know you all would agree with me that every prospect asks how quickly they can get certified. But fast certificates cannot be genuine.

I know with Claude, hitting 5 messages every 5 hours mid-task is a real workflow problem.

Many times, when i used to hit claude limits mid task and switched to ChatGPT thinking the no-limit thing would make up for it. But it's really annoying.

I ended up just being more deliberate about how i use Claude. longer more complete prompts instead of back and forth, which naturally stretches the messages further. not a perfect fix but it changed the experience enough to stick with it.

Finally someone said this! I feel it's not that people are doing less, it's that the feedback loop that used to tell you whether you actually understood something is now gone. you ship, it works, you move on, and three months later you can't explain why you made half the decisions you did.

The bit about taste is what actually stuck with me, though. taste comes from having opinions formed through struggle, not summaries and that's the part AI quietly erodes without you noticing it's happening. Does anyone here agree with me?

The RAG chunking problem is real and most people don't hit it until they're already in production, wondering why the retrieval keeps missing obvious things.

Anyways, will check out the repo, looks like something worth digging into.

oh yesss, except literate programming was still the human explaining intent to other humans. this is more like the human explaining intent to a machine that then explains it back to other humans. hahaha this is actually funny.

yeah temperature is probably worth a run, we noticed even small adjustments changed how the model interpreted formatting expectations quite a bit.

Hey, cool project! We run a bug bounty at Mitigata. If you're looking to add more targets, I'd be happy to connect you with the right person if that's useful.

Have you heard of Mesh Networking between devices?

If enough people in dense urban areas ran something like meshtastic or a similar protocol, you could theoretically pass traffic peer to peer without touching ISP infrastructure at all. range is limited but in a city like Moscow where apartments are packed close together it's at least worth thinking about.

We're sitting around $300/month across a couple of tools. Felt like a lot initially, then we know it's actually worth it