Has a TUI implementation of this approach been considered? a lot of terminal focused situations may not have any GUI functionality but could benefit greatly from the runbook framework.
Are the design specifications, like interconnect, GPUs, CPUs, memory and storage of this new cluster public? I seem to recall xAI has made public theirs. I’m mostly asking out of curiosity and a desire to read up on the SOTA hardware specs being used. EDIT: this has some interesting details — https://www.adwaitx.com/meta-prometheus-ai-cluster/
I assume eventually all this investment should result in price drops for cloud GPU rates. Maybe somebody has setup an automated rate aggregator and collected the data? It would be interesting to see the historical data and monitor the changes, like dollars per TFLOP/hr or something standardized to track over time like other economic data or prices.
EDIT: this is along the lines and pretty interesting — https://www.unitedcompute.ai/gpu-price-tracker
I know I’m mixing two different thoughts but they are connected in my head for entrepreneurs interested in starting independent tech/AI/LLM businesses needing heavy compute infrastructure.
A big problem I find is that if you are a family and have kids you basically have to keep up and that means turning on notifications for messages and emails and then of course that leads to opening the phone, reading email, checking HN (obviously) and then posting a comment on it! urghh
My high schooler is in theatre and they post critical updates via Instagram. It drives me crazy.
And don't get me started on all the custom apps cluttering my phone that these schools and sports leagues get sold on for sharing flyers and other info (Parent Square, Peach Jar, Playmetrics, Mojo, etc.) I guess it's a feature that most of those apps are not well designed and they don't suck you into addictive engagement loops like the big social media platforms.
I don't have kids, so I can't 100% relate, but have you dug into the notifications settings on your phone? They're extremely flexible. I've set mine up so any non-chat notification gets batched into a group that shows at 5 AM and 5 PM. This lets me check on whatever happened the previous evening while I'm having breakfast, and whatever happened during the day after I get home from work. Once I've flipped through all the notifications and done whatever other time wasting things, it goes back in my pocket and largely does not disturb me for another 12 hours.
Maybe something like that could work. If you find there are notifications that are disturbing you, but they really could have waited until the evening, toss 'em in the batch bucket. Eventually you'll tune out all the low-importance stuff and get your life back. Or find some other cadence that works for you. It takes some effort to tune these systems, but I think it's worth it.
I followed the instructions here to cripple my phone using Apple Configurator: https://stopa.io/post/297
Now the browser doesn't work and I can't install new apps. I also turn on "Do Not Disturb" almost all the time, which allows through notifications from exactly 3 people.
I was super-disappointed, attending my recent state townhall meeting, that the only way to participate was scanning a QR code to fill out a survey. After asking for a paper copy (which was never produced), I decided to participate in my own manner:
I stood up and heckled my clown state representatives, for almost an hour, providing audience-appreciated commentary to what I perceive as our failed political system (US bipartisan).
To their toothless grocery sales tax reduction legislation (which'll never pass), I suggested my fellow constituents just shop across the state line, in one of the many nearby grocery stores — just STOP giving our state this money, then maybe they'll consider legislative changes.
Perhaps this fell upon deaf ears, but I wasn't the only audience member frustrated with our legislators' back-patting/inaction. I will vote/shop with my money, elsewhere. I wrote my state officials a letter afterwards, offering common-sense suggestions — hoping this geriatric remembers my participation (he turns 80 soon... just retire already, Congressman!).
Same experience and it’s so bad I sometimes question my sanity. At least I have a notmuch index setup on a server I can fall back on but it’s just so bizarre a “search” company can’t produce an app which can search properly emails!
Is there a good example you could provide of that? I just haven’t seen that personally so I’d be interested in any examples on these current models. I’m sure we all remember in the early days lots of examples of stupidity being posted and it was interesting. It be great if people kept doing that so we could get a better sense of which types of problems they are failing with astounding levels of stupidity on.
One example I ran into recently is asking Gemini CLI to do something that isn't possible: use multiple tokens in a Gemini CLI custom command (https://github.com/google-gemini/gemini-cli/blob/main/docs/c...). It pretended it was possible and came up with a nonsense .toml defining multiple arguments in a way it invented so it couldn't be read, even after multiple rounds of "that doesn't work, Gemini can't load this."
So in any situation where something can't actually be done my assumption is that it's just going to hallucinate a solution.
Has been good for busywork that I know how to do but want to save time on. When I'm directing it, it works well. When I'm asking it to direct me, it's gonna lead me off a cliff if I let it.
I've had every single LLM I tried (Opus, Sonnet, GPT-5-(codex) and Grok light) all tell me that Go embeds[0] support relative paths UPWARDS in the tree.
They all have a very specific misunderstanding. Go embeds _do_ support relative paths like:
//go:embed files/hello.txt
But they DO NOT support any paths with ".." in it
//go:embed ../files/hello.txt
is not correct.
All confidently claimed that .. is correct and will work and tried to make it work multipled different ways until I pointed each to the documentation.
I don’t really find that so surprising or particularly stupid. I was hoping to learn about serious issues with bad logic or reasoning not missing dots on i’s type stuff.
I can’t remember the example but there was another frequent hallucination that people were submitting bug reports that it wasn’t working, so the project looked at it and realized well actually that kinda would make sense and maybe our tool should work like that, and changed the code to work just like the LLM hallucination expected!
Also in general remember human developers hallucinate ALL THE TIME and then realize it or check documentation. So my point is I feel hallucinations are not particularly important or bother me as much as flawed reasoning.
I’m guessing this list is defined by Mac users who all got taught em dash somewhere similar or for similar reasons. It is only easy to use on a Mac. But I wonder what is the 2nd common influence of users using it?
As per the wiki article someone else listed — the compose key was available on keyboards back in the 1980s (notably it was invented only 5 years after the Space Cadet keyboard was invented!).
Some DOS applications did have support for it. The reason it wasn't included is baffling, and it's especially baffling to me that other operating systems never adopted it, simply because
compose a '
is VASTLY more user friendly to type than:
alt-+
1F600
which I have met some windows users who memorize that combo for things like the copyright symbol (which is simply:)
The compose key feels mandatory for anyone who wants to type their native langauge on an US-english layout. The combination[0] is "Compose--." though: –
As it should be. I wish this convention were present across more software, “-“ “- -“ and “- - -“ should be the UI norm for entering proper dashes in text input controls.
Though it does still require nominating a key to map to Compose. And is not generally meaningfully documented. So I’d only call it easy for the sorts of people that care enough to find it.
But then, long before I had a Compose key, in my benighted days of using Windows, I figured out such codes as Alt+0151. 0150, 0151, 0153, 0169, 0176… a surprising number of them I still remember after not having typed them in a dozen years.
In electrical engineering I'm still using a few alt codes daily, like 248 (degree sign), 234 (Omega), 230 (mu), and 241 (plus or minus). I'd love to add 0151 to the repertoire, but I don't want people to think I used AI to write stuff....
My favourite android keyboard has a compose key and also a lot of good defaults in long touch on keys (including en and em under dash). Only downside is last android update causes the keyboard to be overlapped in landscape mode. A problem with a number of alternative keyboards out there.
https://github.com/klausw/hackerskeyboard/issues/957
And whether the user cares to ‘write properly’ to boot. I love using dashes to break up sentences - but I rarely take the time to use the proper dashes, unless I’m writing professionally. I treat capitalization the same way - I rarely capitalize the first letter of a paragraph. I treat ‘rules’ like that as typographic aesthetic design conventions - optional depending on context.
Any pointers to what is SOTA for cluster of hosts with CUDA GPUs but not enough vram for full weights, yet 10Gbit low latency interconnects?
If that problem gets solved, even if for only a batch approach that enables parallel batch inference resulting in high total token/s but low per session, and for bigger models, then it would he a serious game changer for large scale low cost AI automation without billions capex. My intuition says it should be possible, so perhaps someone has done it or started on it already.
