spott

huh?

``` curl -fsSL https://x.ai/cli/install.sh | \ llm -s "Review this shell script and tell me what it does" ```

That isn't unauditable trust...

Is this open source? I can't find a link to download the weights.

Why would they spend a ton of effort ensuring that their customers spend less money on them?

Token economics also are weird. If you design a fancy new frontend that for example uses a cheap model to parse a PDF into text that is fed into an expensive model, you will probably spend more money because you are on API payscale rather than the "max plan" payscale.

You aren't looking for a random set of tokens that have the exact same logit, you are looking for the largest n tokens to have the exact same probability.

This is exceedingly unlikely, as training will only push one of them up for any individual sample. There are likely some pathological situations that could end up with that situation, maybe, but it is pretty unlikely in a general case.

Which products are outstanding in their categories?

Gmail isn't outstanding, search isn't outstanding, maps isn't outstanding.

They are all pretty par for the course. Google used to be outstanding... but I'm not sure of a single product they have that is outstanding (def: significantly better than the competition) anymore. On the other hand I rarely use any google products these days, so maybe I'm not the one to be judging.

Is there a paper on this?

I'm curious how they pre-trained it... I feel like it must have had audio/image output that they chopped off.

I wonder how hard it would be to add it back on.

https://newsletter.maartengrootendorst.com/p/a-visual-guide-... (in a link from here: https://developers.googleblog.com/gemma-4-12b-the-developer-..., which was linked in the text of the post, but not the linkdump at the end).

This is just early fusion basically.

FAIR did this 2 years ago now: https://arxiv.org/abs/2405.09818

I've been waiting for something like this to be released since then.

The annoying thing is that chameleon was multi-modal out based on the same principles, but this model is just inputs... (I'm curious how they did pre-training without having multi-modal outputs as well. I wonder if they just chopped them off rather than support image output).

This could just be because everyone is using direct OpenAI api keys when using OpenAI.

Yea, they are just much higher level programmers… most programmers don’t know the low level syscall apis.

Does ssh need to allow root logins?

Sshing as a regular user and then sudo to root works 95% of the time…

Typo? Or morbid question?

I think of it differently.

Before, you measured diffusion rate of oxygen and inferred oxygen concentration from that (the concentration outside the chamber is always greater than the concentration inside). Dirty membranes etc all changed the rate of diffusion, which caused issues.

After you measure oxygen concentration directly (the concentration inside and outside the chamber are always the same).

This might be it: https://www.eizo.com/products/flexscan/ev2730q/

https://datatau.net used to be kinda that… but it has been dead for a while, and it looks like the spam bots have taken over.

You can. They are offering both options.

Adjusted for inflation?

there wouldn’t be different ideological factions in the first place.

Maybe I’m just very jaded, but I don’t think this is true.

Our values are significantly more aligned than we generally believe, however as long as there is power to be gained by creating the illusion of a difference of values, there will be factions dedicated to ensuring that illusion is maintained.

Multiplex the sensor?

Split the laser into 7 ish spots and read all of them at once?

That doesn’t sound that hard…

This isn’t a horrible idea, but the risk isn’t really big enough to justify introducing that friction.

Cause the risk isn’t actually that bad.

There are three nodes that are running with the same repo. If one of them force pushes, the others have the repo to restore it.

In 6+ months that I’ve had this setup, I’ve never had to deal with that issue.

The convenience of having the agents create their own prs, and evaluate issues, is just too great.

This is the biggest thing I use my Proxmox homelab for.

I have a few VMs that I can rebuild trivially. They only have the relevant repo on them. They basically only run Claude in yolo mode.

I do wish I could use yolo mode, but deny git push or git push —force.

The biggest risk I have using yolo mode is a git push —force to wipe out my remote repo, or a data exfiltration.

I ssh in on my phone/tablet into a tmux session. Each box also has the ability to have an independent environment, which I can access from wherever I’m sshing from.

All in all, I’m pretty happy with the whole situation.

I don’t have the time pressure, so I just use tung oil.

I throw it in a bag and vacuum seal the spoon (with tung oil) for a day or two, then remove, wipe, and let cure for a month.

The resulting finish is largely dishwasher safe for a year or so before I have to reapply. Without the vacuum sealing stage, it doesn’t last as long.

Libghostty is a pretty huge contribution.

2009.

But a classic. Always worth a re-read.

Codex/claude code are the big ones.

Scheduled agents in ChatGPT.

I’ve been using the study mode recently, and that is nice to have.

I want a kagi mcp server I can use with ChatGPT or Claude.

I don’t want to use kagi ultimate (I use too many other features of ChatGPT and Claude), I just want to be able to improve the results of my AI models with kagi.

I think when the half penny was discontinued it had the same buying power as the dime does now or something like that.

So this is long overdue.

Only for C41…. For slide film that sets you back.

I think the problem is that it sounds like you get worse results for slide film with RGB than you get with C41 and white light. So the tradeoff is only worth it if you shoot no slide and C41.

Dynamic range is much better (120dB is ~20 stops, your plustek is ~12 stops.

If that matters, I’m not sure.