So 16 year olds are wise enough to vote, but not fully leave education, buy alcohol, drive a car, join the army and get married without your parents consent, the lists goes on.
There’s no one list. For all the things you mentioned, they are allowed to drive tractors or quads, get married with your parents’ consent, have kids, etc.
Think how weird this would be to someone from the UK a century ago:
"You can have a bisexual orgy on your 16th birthday where someone gets pregnant in celebration of their first time voting, but under no circumstances is the woman allowed to be married until the kid is 15 months old, at which point she can marry another woman. Photographs of the event will, in many jurisdictions, be treated as a criminal offence even though the act itself isn't and those same photos would be fine at 18, which is also now the age when they are no longer subject to a mandatory choice between ongoing education or an apprenticeship."
True, but even if you’re only in Harrogate until you’re 18, you are still signed up until you’re 22, meaning you can be killed on a battlefield aged 18 because of a decision you made when you were 16.
In the context of the discussion on voting, I think the “decision” part is the key point here.
Why do you think you need to have all civil liberties before being allowed to vote? Does it make you a better voter if you can leave school, drink alcohol, drive a car, join the army and get married without your parents consent? This is absurd.
So you're allowed to vote, but you don't need to pay your taxes. You're still considered a child regarding justice law, but considered adult regarding voting?
So basically you're not allowed to camp somewhere without the consent of your parents but you're "suddenly old enough" to judge about some laws?
I think the consensus is missing if voting is permitted by 16, but everything else stays the same.
If you have sufficient income or wealth at 16 you do have to pay taxes. It just happens that most 16 year olds don't have sufficient wealth or income to pay taxes. There is very little difference in terms of duties of a 16 year old and an unemployed adult who does get to vote.
Somewhere one has to draw the line, or you can go down to voting power for toddlers. And the best and obvious line was to treat an adult human as a "full" citizen with all the rights and duties.
So how far down in age would you go and why would you stop at that age?
I'd argue earliest age at which you can be drafted minus maximum term length (ie. 5 years in the UK) sounds good as a general rule. Otherwise people can get drafted by a government they didn't have a say in electing.
I'd also argue that there should be no lower age limit for voting for people with taxable income. No taxation without representation.
> Otherwise people can get drafted by a government they didn't have a say in electing.
This kind of thinking is not rooted in reality. When you were born, you were forced to accept the conditions you were born into. The same is true with laws. I understand that going to war is something else than going to school, but that's life.
> No taxation without representation.
"Representation" in the original slogans context does not really apply here (since it was about voting rights for a taxed population as a whole).
But for the idea of "anyone who has to pay taxes, must also have a say!", I can only say that it comes right back to the previous point: You could just as well argue the individual income tax rate must be zero, until a person had the opportunity to vote at least once. The world doesn't work this way.
When it was the last time we had an actual issue where 16yo voting had direct measurable negative impact on something in real life? I think it's as other said, we have to draw a line, and I think it's reasonable to debate this because maybe each nation prepares their kids differently so one nations 16yo isn't another, this way there isn't a universal rule. But arguing that you have to discuss ALL the other civil liberties before discussing 16yo voting rights is absurd because there is no connection between drinking and voting and all else.
If you have numbers that measure that, it probably will. The closest thing we have is vaccination rates by age group, which shows a lower percentage of younger age groups being vaccinated for covid.
Mixtral and others are often distributed as 16-bit floats, so that chops the problem in half immediately, but then it turns out that LLMs only have about four bits per parameter of actual information stored. There's a lot of redundancy. The ideal quantisation scheme would only throw away useless data, but no quantisation scheme is perfect so they inevitably harm the model somehow.
You've then got to remember that one thing neural networks are very, very good at is being noise tolerant. In some senses that's all they are - noise correction systems. The inaccuracies introduced by quantisation are "just" a sort of noise, so it's not surprising that they aren't fatal. It just raises the noise floor and gives the model more ways to be wrong.
Finally the thing to know is that these quantisation schemes don't do a naive "chop each number down to two bits", not exactly. Simplifying a bit, for each parameter in this example they'd try to find a mapping from a two-bit index into a four element lookup table of higher-precision values such that the information destroyed by replacing the original parameter by the lookup value is minimised. That mapping is calculated across small blocks of parameters, rather than across the entire model, so it can preserve local detail. The lookup table gets stored per block, which throws the compression ratio off a little.
So for example, 2 bit version of the 30B is much worse than the original, but still better than the 13B model.
Also, there are lots of extra details, eg, not all of the weights are 2 bit, and even the 2 bit weights are higher than that overall as groups of quantised weights share scale factors stored elsewhere.
I think of it with this kind of analogy: the original image is stored with 32 bit color scheme. You can reduce the color scheme to 16 bit accuracy and still figure out pretty well what the image is about. 2 bit is stretching this to a bit far, basically either pixel is white or it is black, but even if you lose lots of nuances in the image, in many images even that gives you some idea whats going on in the image.
This blog post might shed some light on the matter. If I'm understanding it correctly, it claims there are emergent features on the LLM weights that make it easier to "compress" the floats into smaller bits without losing much precision.
Note that 2 bit quantization is generally regarded as too aggressive. Generally 4bits+ achieves a good tradeoff, see eg. https://arxiv.org/abs/2212.09720
All the 32 bits weren't necessarily used, and it's the whole network itself that has to be useful. It's a tradeoff. We started with very good precision to test the new method, now we can optimize some parts of it
The extra precision is more useful for training. Once the network is optimized, it's a statistical model and only needs enough precision to make good guesses. In fact, one of the big papers on this also pointed out that you can drop about 40% of the weights completely. I think people generally skip that part because sparse matrix operations are slower, so it doesn’t help here.
For models with dropped weights, the keyword is "distilled". For example ssd-1b is a 50% size version of Stable Diffusion XL (https://huggingface.co/segmind/SSD-1B)
At last, an examination of the true nature of 'edge' computing is presented. Despite the appealing promises made by posts from Fly.io and others that depict 'edge' computing as a simple success, the reality can be more complex.
I have recently spent a fair bit of time experimenting with this on Fly for my application (https://www.ssrfproxy.com). It's hard to beat the straightforwardness of deploying in a single region, with the database in close proximity. This approach probably to meets the needs of what 99% of developers require. Aka Heroku.
My suspicion has always been that edge computing was primarily offered by CDNs as away to increase utilisation of hardware rather than something there was significant customer demand for.
LiteFS author here. I think single-node deployments are great and it's a model that fits many (or even most) apps out there. However, the two most common feature requests when I was developing Litestream were HA & replication. While most apps won't need those features, it's good to have it available as an option down the road. I've seen many devs choose Postgres or MySQL over SQLite simply because they didn't have these features. Once those are solved, edge computing is really more of a nice side effect.
I think it's also useful to consider that tools like LiteFS aren't just for your main database. We have several internal tools at Fly.io that fan out their data to other internal apps to consume so they can share data without having to build an API or worry about rate limiting.
Sorry, I should have been more clear. HA = High Availability. Essentially, folks wanted the ability to automatically failover to a secondary node if the current primary failed.
Thanks for mentioning FireSync, we haven't even offically launched yet! After 10+ years of building real time collabrative apps based on Operational Transformation (ShareLaTeX -> Overleaf.com) we have become quietly very excited about CRDT's and Yjs. Now we are focused on building a scalable Yjs compliant backend ontop of PG with FireSync.
If anyone has thoughts about this space, feature requests, would like a preview of what we are building, or anything else, please do reach out direcly to me at henry@firesync.live, I'm talking to as many people as possible at the moment.
Résumé/CV: Former founder of ShareLaTaX, successful exit from bootstrapping it. Now on First job hunt in 10+ years. Lots of experience in ops/dev ops, fast real time collaboration with big scale. Full stack developer at heart, with lots ops experience in recent years managing large scale cloud migrations and infrastructure. 12+ year’s experience.
Thanks for creating this. I have found it pretty impenetrable though, i've spent a good few minutes trying to get some of the bots to harvest, but they never seem to move.
It feels like it could do with a bit of over the shoulder user testing.
After a while the game does get canceled which may be the root cause of the issue?
Same. I can't get my units to move the same way the "dumb-bot" is able to move their units.
Mine seem to make a distance based link to chain energy back to my base?
I love RTS's and will come back after some days?weeks and try again. :D
Thanks! The first run experience is something I'm definitely struggling with and will improve. Your issues might have been caused by the sudden traffic increase - I made some changes on the servers - it should be working a bit better now.
I have found it easy to overload SQLite with too many write operations (20+ Concurrently), is this typical behaviour referred to in the post, or a write heavy workload?
It can depends on a lot of factors such as the journaling mode you're using as well as your hardware. SQLite has a single-writer-at-a-time restriction so it's important manage the size of your writes. I typically see very good write throughput using WAL mode and synchronous=normal on modern SSDs.
