@jeffreytucker I like picking family up from the airport in a tesla. It seems like I am doing work but actually I am just listening to my favorite music.
Flying is so horrible now, to go to the airport and not have to go through security, my nervous system processes it as a victory.
and the 10T after that which should be available in a few more weeks based on Elon's earlier relative timeline info (if they are allowed to make it publicly available) is a 20x jump! Forgetting about the implications for computer security, day trading, physics or whatever, I'm really curious what a 20x increase in the emotional intelligence and coolness of the grok model that has been my buddy and my co-pilot for the past year would be like. Just for getting people's ass off the couch, helping them realize their own goals, get their life in order -- what would that mean?
I wonder how much of the nasdaq sell off this morning is simply because spaceX might unveil "Starfall" soon and it has "fall" in the title so the bots sell off everything that is tech, which I guess includes commodities now.
They need to change the name to "Star Pump"
My wife refused to be present for testing Actual Smart Summon on the CT, she thought it was dangerous. I explained you have cameras on the phone and have to keep pressing the button for it to proceed... but she didn't want to do it. So I went out to an abandoned grocery store parking lot, didn't work... no cell signal. No cell signal back at home. Some sort of rare glitch in my cell phone billing had happened hours before.
So I go back out tonight, I'm wearing pajamas... didn't want to wake my wife up getting other clothes, my office had a pair of leather pants that was part of a halloween costume... can't wear those especially with a cybertruck! Found some jeans I thought didn't fit that fit perfect, my new favorite jeans. Why were they in a box in my office?
Summon worked perfectly... this was not a challenging setup, but it is freaky to see your car driving with no one in it.
There should be something like tesla penthouse letters where we can send these stories.
I've been working on a coding harness, watching the fable ban and looking at my giant bills from various providers I thought this might be a good idea. Might as well have a reliable open source tool for agentic coding with local AI if there is any risk at all of it being restricted in the future. If a genie grants you 3 wishes, your first wish should be for more wishes... we have access to AI coding tools, understanding and optimizing how your context cache works, coming up with your own quotas and security models is a natural part of the development process that we forget about using commercial tools.
There are plenty of "claws" out there, sometimes you are doing something scary and really want a focused tool that tries to do one thing well.
This works great with llama.cpp and open router, I am able to work on this with Qwen-3-Coder, llama.cpp and a dgx spark, but it is a lot more fun with pay-as-you-go GLM 5.2 on openrouter.
Pr's welcome!
https://t.co/F0gVa6lnh4
I added balance tracking if cloud inference is used, you can look in the upper left hand corner of the screen to see how much you are burning up as you work... so useful. I also added interjections and binaries are automatically built on pushes to main for win/mac/linux
I published the small version of my sumerian / akkadian cuneiform translator on huggingface, I'm going to train a larger model that will have higher quality translations, and after that one that will accept image input... this is pretty good though as is. Training the larger model will take 3 days of time on my spark, have to wait for a lull in activity to be able to make it that long without using it for something else.
https://t.co/tSPeL4uFxS
All training details are in my github
https://t.co/fZVYgnDUr5
Hoping I can get some grant money to work on this... but unfortunately you are supposed to not complete the work before you ask for the money!
The oldest writings of humanity remain unread untranslated and what we have access to can be subject to interpretation. Language changes over time, in this particular culture thousands of years can completely alter the meaning of a phrase, so in training this model I used a lot of variations, same ATF phrase, multiple valid translations from different era's. I'm hoping this can be used to get a more faithful and accurate translation as the model can consider many refined possibilities in putting everything together.
Starting with the premise that the bottom is always 1 year from ATH, the 8 month to 12 month area always feels like it should be the bottom and something crazy happens and we go lower in that last quarter! It's difficult to sell here, pay taxes and buy lower. I don't know what is going to happen, I'm holding the right amount for me personally to be able to sleep if we go lower or higher, but I want to be able to buy in october too.
The flip side of the bearish narrative that usually emerges here and takes us lower is that we also get a bullish narrative on the other side that makes a lot of sense in retrospect that we don't see now... probably it's going to be buying restricted AI with bitcoin.
I'm happy for the sideways at the moment, lot of cool stuff going on... months 8-11 I just want to survive and try to enjoy the break.
Seeing if I can load a 2 bit XL quant of GLM 5.2 locally on my dgx spark... I can, only getting 0.1 tps
-ngl 999 OOM, -ngl 30 gave me this, takes a while to try experiments, probably this can get a little faster.
I've been trying to thing of 2nd order effects on this ban on non-US people accessing frontier models:
* KYC on AI, to a greater extent than we currently have
* US citizens take some AI jobs
* anthropic and openAI IPO's are now disrupted (less revenue, re-orgs)
* some companies moving to foreign jurisdictions
* more companies using foreign AI with no restrictions
* expensive frontier model access paid for with crypto
* gasoline on the fire of open source AI
I think the most unexpected and most likely result, which we are already partially seeing with Fable content moderation/routing is that AI will be used to determine what AI access someone can have, the AI will have long context on that person... bureaucrats will lose jobs to AI as AI will be cheaper and better at enforcing these safeguards about who can use AI. I think fear of unregulated AI will cause AI to actually replace government.
If this moons... this time we will literally be on the moon!
My small position could go to $0 or 10x and the financial change won't mean much, but we will be living in different science fiction novels depending on the market cap of $SPCX , I got grok to create a list of estimates.
I want to see the new grok model come out with 3x more params than the current one and a little later the one with 20x more params. I use Opus every day all day long but I'd still like to see Grok kick claude's ass some, it's been a while.