LLM are just text predictions based on what people would say in available digital works (like comments). Its honestly a fascinating glimpse in online sociology.
This would be like playing DnD where you see a painting and describe what you would do next as if you were the painting and they an artists painted the next scene for you.
The artists isn't rolling dice, following the rule book, or any actual game elements they ate just painting based on the last painting and your description of the next.
Its incredibly nove approchl if not obviously a toy problem.
Honestly I thinkyour self driving example is something this could be really cool for. If the generation can exceed real time (I.e. 20 secs of future image prediction can happen in under 20 secs) then you can preemptively react with the self driving model and cache the results.
If the compute costs can be managed maybe even run multiple models against each other to develop an array likely branch predictions (you know what I turned left)
Its even cooler that player input helps predict the next image.
Reminds of an observation my dad made once about how neighborhoods are named after what had to cleared away to build there. Oakwoods. Rolling hills. Pleasent view. Etc etc
Religion is an opiate. The best way to reduce its abuse is by addressing the underlying pain. When people conditions get worse they look to things to help numb the pain.
Is there any objective measured ways or at least subject reviews based metrics for a model on g8ve problem set? I know the white papers tend to include it and sometimes the git repos, but I don't see that info when searching through ollama for example.
I saw you other post about ollama alts and the concurrency mention in one of the projects README sounds promising.
Very much pro Open Source AI. Especially as a concept digital public good. With https://petals.dev/ being the most promising option that regard (imagine something like RAG for the arch wiki with very large models supported by the community!).
It feel very enthusiasts right now. Where I feel like I'm just on the cusp of having usable set up.
I personally really want a full Dev that just takes gitlab issues and runs codes against tests until it passes, and then cycles between attempting to explain what it doing and refactoring until that explanation is reasonably simple, then submit PR.
At the moment I am trying to use it as a copilot (ollama lama3, continue, and devonAI vscode plugins) all on my MacBook (my Linux machine were too small gpu wise, at least first time I attempted). That said it ok for questions no real luck on a decent experience for actually making anything.
The next step to me for it to move from enthusiast to hobbiest would be:
Models that just work on my machine. I had to do a lot of trial and error just get performant models.
Models just my use case. I don't know what model support tooling, or multimodal inputs. What models are actually optimized for programing, for actions (ala openinterpretor), for reviewing documents, etc.
For federated (like pedals.dev) I feel like I need some sane data guardrails. I don't want my medical documents anywhere near "bittorrent style" anything, but would absolutely love to leverage it for better outcome on opensource projects without secrets file. This also feeds into point 2 to me.
More sane RAG. Maybe even IPFS links to caches or DBs for popular data sources (like code docs for example).
I feel like there has to be a better way for this. Maybe its just selinux rules for data tags for locking down my local system and some routing config file at the root of my projects. Idk tbh
Win 7 was cool to tinker with but buggy as hell to the point Ubuntu was just as much work as the time but with even more possibility.
8 was so bad it was worth skipping.
10 was the peak of it just works for windows. Gone were the days of troubleshooting triple AAA games on my PC, they worked or needed patched by the devs.
That said if I NEEDED something to work choosing windows 10 or server was an exercise is maschocism. Need this container? Unsupported. Need this service configured like this? Gfl finding where that is set. Need HA? Just Ha. Certain network configs beyond basic client? The guys with decades of Windows admin exp still have no idea.
I had to troubleshoot both but windows gives you the nice feeling of being able to say "this sucks they should fix that" because I know ain't. Its not built for me to fix it either. Linux however begs you too. Its all there, you can do anything, even you might not really want to.
I air out my mattress when I wash my sheets (leaving it uncovered for that day) and push it vertical on the wall (like a Murphy bed).
I know I'm saying don't buy anything, but I have always gotten those terry cloth textured water proof covers for my matteress too. Getting fluids out of mattress is just a losing battle
But why do we really need convenience, right? I agree there is a real argument for outsourcing labor to a more specialized group, but honestly a lot of the time we trying to save hours or minutes so we can spend more time and money consuming some bs to fight our boredom.
Tbh about everything. Most of retail is just an industrial scale of the addage "a fool and his money will soon be parted".
Buy second hand, its fine. You probally can figure it out yourself, try to diy. Look at what people are actually doing not the brand of tool they are doing it with. Its a saw, you saw with it, you can get away with sawing a lot of stuff with the same cheap saw.
Soaps are just collections of chemicals, powerwash for example is just dish soap plus water and isopropal alcohol.
You can probably cook it at home. It will probably be better and better for you, because a pound of lard or cup or sugar looks like the red flag it is when you go to cook with it.
Your bed might be better on the floor, then on a frame.
You are probably better off walking or biking then driving.
You probably don't need to watch more shows anyways so why get fleeced to subscriptions.
You probably don't need to play games as much so you can pass on that game. You probally don't need to go out for a drink. You probally don't need to go out for a meal. Etc etc
Honestly, I'm a hypocritical ass saying some of this, but its true. The urge to go spend spend spend, isn't a fluke its just successful sociol engineering to separate us fools from our money.
Because not enough creative believe themaxisnm of "fuck you pay me".