Local AI Alex

Chem engineer / chemist & founder. Running open models on 4× DGX Spark (will be 8 soon). Writing on local AI: real industrial use & my occasional perspective

[Local AI Alex @AI__Alex]

https://t.co/gJZXMgLJ4B

[Local AI Alex @AI__Alex]

This is a formula for success in multiple paths in life. Relationships are one of the defining features of my own business. Long-term customers, long-term suppliers, long-term relationships with other providers. Would never undercut any of my relationships for any short term profit or to optimise the value I can extract out of them. We all win together and we keep doing this over again

[Local AI Alex @AI__Alex]

@niklaslenz_ai I pretty sure both of the mikrotik switches suitable for Spark deployment were globally sold out a month or two ago. Not sure about lead times to buy one. The other alternative is a $20000 minimum NVIDIA switch for enterprise scale hardware

[Local AI Alex @AI__Alex]

@KyleBehrend Shh 🤫 IoT store was my secret store that I told you about in confidence

[Local AI Alex @AI__Alex]

@KyleBehrend Shh 🤫 posts like this will surely exhaust the remaining sparks on the Australian market in no time. What will I do if I need to buy a couple more sometime?

[Local AI Alex @AI__Alex]

@TechMDAI Tell me about this product, why you bought it / wanted it so bad, what it's now going to allow you to do... I'm very interested

[Local AI Alex @AI__Alex]

@MichaelGannotti So what happened?

[Local AI Alex @AI__Alex]

Chris, local AI is where you want to be if you're serious about AI. I still have frontier subscriptions but the local AI technology is at the stage where it is possible to get almost the exact same level of intelligence as you would from anthropic or open AI. Aside from this fact there are serious advantages to having your own private ai at home. For example; (1) absolute privacy - you can ask it anything you want and none of your private information is shared externally, (2) you can download uncensored models that have had their woke filters removed, and three models will answer any questions you ask, which is very different to online models, (3) the performance never degrades - you can instruct your model to work on a task for as long as you want so that the quality of the response it gives is exceptionally high - online models get shittier and shittier over time and will try to short change you by cutting off it's effort as early as possible to save money, (4) lower cost for power users - you only pay for electricity not tokens, (5) infinity customisable whereas cloud AI is very restricted on how you can customise it, (6) extremely quick instant resposes with no latency so long as your hardware is adequate and configured properly, (7) no risk of the vendor stealing your IP and ideas, (8) cloud AI is constantly profiling you and this can affect the level of service it provides to you - if it doesn't like you it will degrade the quality of the results it returns to you - if it's really doesn't like you then it might even cut you off ... (9) Cloud AI is controlled by someone else and the people controlling it can use it to force their agendas onto you (10) the government is also in this cloud AI space now and you know how governments like to control you and decide what you can or can't do and what you are not allowed to know and what propaganda you have to eat.... There are many more but this should be a good enough start for you to understand why local AI is better.

[Local AI Alex @AI__Alex]

@superalesha @MiaAI_lab This is a really awesome video Alexey. It was put together with the message so well.

[Local AI Alex @AI__Alex]

Being right in a Git log and being understood by an audience are now two different jobs. A scene that only does the first will keep winning arguments and losing the room.
I’ve laid out some potential path-forward standards and tagged @Blackwellboy and @MiaAI_lab to keep it entirely fair. Read the full piece here: https://t.co/b7OSnjw2d2

[Local AI Alex @AI__Alex]

https://t.co/gJZXMgLJ4B

[Local AI Alex @AI__Alex]

The controversy surrounding @MiaAI_lab and the exposé by @Blackwellboy has split the local AI timeline in two. One side sees an open-and-shut attribution case. The other sees a pile-on.
They are reacting to two different things.
Here is my take on the macro-shift happening right now: https://t.co/b7OSnjw2d2

[Local AI Alex @AI__Alex]

https://t.co/gJZXMgLJ4B

[Local AI Alex @AI__Alex]

As a local model user who also runs a traditional engineering business, I live in the gap between technical truth and customer perception every day.
Local AI is running into a classic culture clash: a 3-year-old scene operating on decades-old hacker norms colliding head-first with the mainstream creator economy.

[Local AI Alex @AI__Alex]

@Blackwellboy You pushed me to think harder about the builder side, so I wrote it up properly. Your five open questions are all about credit, which is the part receipts can settle. Tell me if I've misstated you anywhere. https://t.co/Q8wwS7ub8e

[Local AI Alex @AI__Alex]

https://t.co/gJZXMgLJ4B

[Local AI Alex @AI__Alex]

@MiaAI_lab Good to see the identity question answered directly. The credit questions are checkable, so a public changelog of what was fixed, with the commits, would close them faster than any thread. I wrote up why the two keep getting tangled: https://t.co/Q8wwS7ub8e

[Local AI Alex @AI__Alex]

https://t.co/gJZXMgLJ4B

[Local AI Alex @AI__Alex]

@KyleBehrend Shh 🤫 I might need to get one or two more

[Local AI Alex @AI__Alex]

@MiaAI_lab It's unfortunate that sometimes things like this happen. Take it as an opportunity to reflect and understand that we are different but our purpose is the same. Local AI available for everybody 🙌

[Local AI Alex @AI__Alex]

2 from "IoT Store" they apparently have a shop in Kewdale WA and their website said they had stock, and 2 from PLE Computers. I deliberately didn't order them all from one place just in case I get a cancelled order. There are a few other places advertising them online and them stating that they do in fact have stock on hand. I would avoid trying to buy from anyone who doesn't explicitly state this. Do a google shopping search for DGX Spark in Australia, weed out the dodgy looking results. Try to buy from the most reputable vendors even if they are a little bit more expensive as advertised.

[Local AI Alex @AI__Alex]

I was holding out for a m5 ultra 512 GB instead, but recent comparison test results performed by @MiaAI_lab, plus the disappearance of Sparks in the market, plus other cost and availability escalations of late caused enough fomo to force me to buy 4 more. There are still some Sparks available online in Australia (not many) so I concluded that if I didn't buy the extra hardware capability/capacity that I have wanted for month's right now I would regret it badly.