mv4

It's quite shocking to me how many experienced, tech-savvy people, who used to care about cookies and ad tracking - are now willingly sending their business strategies, highly confidential contracts, and intimate personal issues to a cloud provider because "it is only $0.0x per million tokens!".

I remember when I was working at Meta, I was offered a "high impact" opportunity to work on youth products. Nope.

Hermes is the way to go. Accounts provided by the frontier labs are just too volatile.

It 5..10 lines of code. :)

That's correct, and I was specifically referring to the example chosen - where scale and perspective are known. Template rotation is relatively easy as well - but partial obstructions would pose a problem.

Another application where template matching would work brilliantly? Car counting in parking lots using satellite imagery.

Source: I did this [1] using OpenCV and template matching. Outperformed "Cars Overhead with Context" models.

https://abcnews.com/International/satellite-data-suggests-co...

Exactly my point. Template rotation is a trivial operation as well.

Ironically, the pill counting example selected to showcase "the best vision model" can be easily solved with OpenCV template matching, a technology created 25 years ago.

It's never 0% though. You pay a transfer fee (e.g. 5%). The 0% promo period is also limited.

The consumer is screwed either way, whether they try to "transfer" balances or not.

Technically, this can happen. Politically, this won't be allowed to happen - just like digital media ownership never happened.

The frontier labs will do their best to prevent this from happening. Their financial model won't work if people start running open-weight models on their local hardware.

This is why banning Chinese open-weight AI models is a major policy debate in Washington. The labs can't survive log-term without subsidies, and a ban can act as a subsidy.

This is giving me Backrooms vibes!

My personal definition is if you can't resell it, you don't own it.

I've been using MiniMax M2.7 with vllm on my dual Nvidia Spark cluster. Slow (<20 tps) but functional for most of my use cases.

Translation: now I can FIRE.

What the hell did I just read.

Intel has already tried this (CPU+GPU+NPU) in partnership with Asus [1]. Big announcement at CES back in January. Not seeing much traction. Let's see how this one is received.

[1] https://www.intel.com/content/www/us/en/ai-pc/overview.html

I am sure it can. It will still generate a lot of heat when under load.

Are you telling me I should go for it? :)

I do have a dual DGX Spark cluster running MiniMax M2.7 already so I am all for on-prem. But will be interesting how this old machine will perform!

I have an old 192GB DDR4 Dell Precision with dual Intel Xeon Gold 6130 that I've considered spinning up. What's giving me pause is 250W at idle.

You just described the absolute nightmare scenario for the newly minted trillion-dollar companies whose only hope is for enterprises and SMB to move all their business processes to the cloud, with employees competing at token maxxing.

Minimax M2.7.

I configured a dual DGX Spark cluster, and it's certainly "good enough" for my agentic and coding needs.

If people figure out how to run agents on-prem (already becoming feasible for both agentic tasks and coding on consumer hardware like Mac Studio 128GB+ or DGX Spark with some models) these companies will be in deep trouble.

Privacy is also a huge issue.

Only took 19 years!

Can someone recommend an IDE that can be used with a self-hosted model (via OpenAI or similar)?

I mentioned the same thing and it's getting downvoted lol

Looking at the comments here, it appears people are trying to do too much, too soon, which inevitably backfires. The key is to put OC in a sandbox (don't give access to your real accounts, it must have ints own separate non-admin accounts for everything only "invite" with limited access as you would a contractor) and generally treat it as a new employee during a trial period. You'll be surprised by how effective it can be if you pace yourself.

As for "you can easily do X,Y,Z with a cron job" we tech people often underestimate what hiding the complexity could do for UX.

"For a Linux user, you can already build such a system yourself quite trivially by getting an FTP account, mounting it locally with curlftpfs, and then using SVN or CVS on the mounted filesystem." -- top comment on "Show HN, Dropbox (2007)

#2 also provides an opinion on why a certain development is significant and how it relates so something else I am working on for that client.

The process is definitely more pleasant to me than setting up cron jobs and scripting things. I have a business to run.

No recurring expense, just the hardware upfront cost. My next step is run the models locally as well. I used to work at FAANG (incl. compliance) and I would never use a cloud-based assistant of any kind.

It is buggy. It's also good at self-diagnostics and I was able to either fix things or find working workarounds by collaborating with it.

Operations: mostly tasks and reminders, I often think of something while on the road and send a voice message (like I would to a real PA). I use Kokoro-based TTS for local text-to-speech. Anything from "remind me to discuss X with so-and-so tomorrow" to "I have a multiple PoC projects w/ large entities starting next week - think about how I can best handle that" to "brainstorm how can I maximize upfront revenue with them", etc

Sales: using Google/gemini web search API and it's best run off-peak due to rate limits 503 Service Unavailable (everyone is overbooked when it comes to AI) to see what's happening in the space, any new developments involving companies I care about - and send me a daily digest with an overview and conversation topics.