Victor Valentine Romo 🌴

Introducing my first FOSS dupe: Declip. 🙂 Local filler removal for talking-head video. No cloud, no subscription, no upload. Runs on Apple Silicon. Replaces Descript's filler word removal + Studio Sound with a single Python script using Whisper (MLX) and ffmpeg.

@AravSrinivas so it will actually do something automated unlike Comet for Android?

We're deepening our partnership with Samsung by powering the AI on their browser that's pre-installed on 1B+ Samsung devices (100m+ active users), extending the partnership that allows Perplexity to power Bixby and be pre-loaded on all Galaxy S26 devices with alongside Gemini.

@elonmusk Not as troubling as the goonware you're making

Troubling

the only time i'm pushing limits is in major restructuring of my vaults, bots, or websites. this is maybe a once a week thing. this upper limit has only been reached by me when i displatched 50-agent swarms to take a domain, build the semantic map, research market sentiment, design a brand, draw up wireframes, build the actual site, outline 300+ web docuemtns, write them, interlink everything semantically, "SEO" it (technical, on-page optimization), then launching to the internet so a 'single' push of the above process (which is technically a one-shot thanks to my vault construction) will burn something like 10-20% of a weekly allotment and certainly fills a 5-hour limit in less than 30 minutes. but that's to run MY process for anyone who's focus is to speed up whatever-operations their doing, I don't see too many use cases where these limits are impacting them (on Max 20x) However, I can imagine for heavy-engineering or programming, these limits would be crushing capabilities during the most crucial hours of the day

best part of the new claude limits is that i'm in the Eastern timezone and can get all my shit done before the end of GMT's workday and the start of PDT's workday lmfao

We have evaluated the Anthropic leakers Capybara 10T beta AI model and it is better, but the scaling laws on the LLM system they use has not been improved, it is suggested it costs $10 Billion per year to run. Next Token will be replaced by Next Concept. We in garages are testing Next Concept based AI and it has far longer scaling laws. Yep you heard it here first. Bookmark it.

A business that doesn't seek to codify their practices into the internet corpus will soon no longer have the opportunity to.

It should be obvious that humans seek connection (it's the medium of energy transport)

LOL

If you listen hard enough, you can 'feel' everything growing together. We are all branches of the same tree.

Every single prospect i talked to in the last 6 months asks when the next one is coming. I say March/April. It never misses.

"i'm a pro" and yet

@BenjaminBadejo yes i will, because i planned for it aifirstsearch.com

When Anthropic rug-pulls you by ratcheting down its unquantified rate limits (like it did this week to thousands of users), you will lose all of your work when you sign into another subscription, or have to switch to API billing. API billing costs orders of magnitude more. Getting stuck in their product ecosystem means they can hold your data and workflows hostage and rob you blind — and you will have no choice in the matter.

building ai product is closer to tuning a musical instrument than writing software.

@DoctorPerin Is this CBT?

@b2bvic Steve! Boom, named.

you couldn't name your thought patterns if you knew the rhythm

this guy literally called this 11 days ago and this was thariq's response lol

There will be more discourse in the coming months and years about AIs needing to be "safe and effective." The inversion is happening now; again.

Stop parading the inevitable

@notaedonis the "It's not this, it's this" rhythm is one i hear a LOT on older TV shows now that I know to listen for it, you'd be surprised how many of those cadences are exactly the same. Wherever it ends up, i don't think it will be good lol (for most)

@b2bvic thats interesting I never thought about it like that, I always likened it to like newscaster or lecturer type speak. like a lot of times it seems to come off as really condescending & matter-of-factly even if it isnt meant to be that way crazy to think where this will all end up

I feel like i'm going crazy with how often i'm able to pick out people writing tweets and messages with chatgpt now. it feels like i'm in some type of psychosis - that's not normal, that's a sign of a bigger issue - one that may need to be addressed.

@FlyInTheCocoa I have two repos to address this point specifically, one is the vault build itself (subtlebodhi) the other is a prompt to prevent sycophancy (observer protocol) Happy to answer any question you run into along the way

you're bitching about AI i am acitvely pirating ideas because there moat is your own effort and creativity