I don’t believe this is a meaningful distinction when we’re not going to agree on how to judge performance of software engineers. If this were solely about income, it might be an important distinction.
The article assumes a normal distribution, making the distinction moot
But it is useful to question whether that is true in all cases. The cases that aren't normal-distributed might be exactly the cases where it pays off to be neither average or median
there is a major shortcoming in this assumption; everything we've seen related to the internet and technology in general suggests there is rarely a normal distribution. I think it's way more valuable ato frame the questions as a long tail (pareto) distribution and a "good enough" cut-off point.
I disagree, I think the bigger issue is blanket banning IPs because they can't decrypt the traffic.
This is the kind of manufacturing consent that would make some people be in favor of the government MITMing crypto so that they can verify that I'm not doing something naughty.
I mean, there are still philosophers metaphorically fist fighting about this stuff. Last time I stepped into the fray on this topic I got clapped back by someone from an area of philosophy of mind from after I graduated. It was an interesting perspective that was unaware of, but I studied language, not mind:
I mean mathematically you need at least one vector to propagate through the network, don't you? That would be a one hot encoding of the starting token. Actually interesting to think about what happens if you make that vector zero everywhere.
In the matmul, it'd just zero out all parameters. In older models, you'd still have bias vectors but I think recent models don't use those anymore. So the output would be zero probability for each token, if I'm not mistaken.
To switch around quick. So on my system I've got Super-{1-9} for workspaces, Alt-{1-9} for tmux panes.
Also if you want a vi-like copy mode (where you can select and copy stuff) that opens using Alt-/:
bind-key -n M-/ copy-mode
set -g status-keys vi
set-window-option -g mode-keys vi
# v to trigger selection
bind-key -T copy-mode-vi v send-keys -X begin-selection
# wl-copy if you use wayland
# mouse selection in copy mode to copy
bind-key -T copy-mode-vi MouseDragEnd1Pane send -X copy-pipe-and-cancel "wl-copy"
# y to copy
bind-key -T copy-mode-vi y send-keys -X copy-pipe-and-cancel "wl-copy"
And btw, Ctrl-Shift-v to paste system clipboard
You can also put the config into ~/.config/tmux/tmux.conf
edit: And if you want a powerline-style design:
set -g pane-border-style "fg=colour252"
set -g pane-active-border-style "fg=colour25"
set -g status-style "bg=default,fg=default"
set -g status-left "#[fg=colour231,bg=colour25,bold] #S #[fg=colour25,bg=default,nobold]"
set -g status-right "#[fg=colour67,bg=default]#[fg=colour231,bg=colour67] %Y-%m-%d %H:%M #[fg=colour25,bg=colour67]#[fg=colour231,bg=colour25,bold] #h "
setw -g window-status-format "#[fg=colour243,bg=default] #I #W "
setw -g window-status-current-format "#[fg=colour28,bg=default,reverse]#[fg=colour231,bg=colour28,bold,noreverse] #I #W #[fg=colour28,bg=default,noreverse]"
There's this sentiment in Germany that if you can't make in industry, you work for the government or - even worse - become a politician. It seems like Mistral took that to the next level; they can't compete so they do lobbying instead.
Being European, I love the idea of European AI labs. But I wish there was more competition.
That being said, as a German for example, I can't think of an AI company successfully training a competitive foundation model here. The copyright mafia would take your investor's money before you could even finish the first training run (hyperbole.)
Z.ai, Moonshot, Deepseek - which of these spent trillions? Their best models are much stronger than Mistral's, the first two have models that are competitive with the US frontier ones.
Not hating on Mistral here, their latest Mistral Small 4 is quite capable and a very reasonable choice for its cost and size.
Is really a compete XOR lobby the government? The AI companies in the US do both, and nobody says "OpenAI can't compete so they do lobbying" or "Anthropoic can't compete so they do lobbying".
In that case, hopefully the copyright mafia will take the money from US and Chinese LLM companies and redistribute it to the people who did the actual work fueling the models, such as myself.
I did not spend 10 years writing (A)GPL code for all of it to be stripped of its license, remixed and sold for profit.
Of course in a truly just world, the LLM companies who took my code without permission would beg with offers of owning a share of them because if I didn't consent their models would have to be destroyed.
Since my work is apparently so valuable that they just have to have it, it should count towards my retirement age too.
But it begs the question why you chose (A)GPL instead of a permissive license in the first place. If you are OK with people training and using LLMs on top of your work not giving users the right to inspect and modify, you must have logically been OK with it before. The existence of LLMs or even AI is irrelevant to your position.
That's my point. You don't care about freedom or user rights and you never did. LLMs don't change that, they just give people like you a way to (for now) legally take without giving back and without respecting the wishes of the people whose work you're building on.
Look at consent in sex and how long it took to make everyone accept it's something you need to have sex with someone. Some places are still not there yet.
I firmly believe using other people's work should require the same level of consent.
> Can you prove your work was used in any of these models?
They admit it themselves. We also know how aggressively they scrape everything they can get their hands on because projects like Anubis[0] exist
> And if so what percentage of your work constitutes the model?
That should absolutely be quantified, yes. My part is tiny but together with other people whose work was taken without consent, we make the vast majority. Last time I napkinned the math, I estimated making the models took 10^12 hours of work (nearly all being scraped public and possibly even private projects), out of which only 10^6 was paid (work by the employees of the LLM companies). So roughly 10^12 remains unpaid.
If you think the belief that people should be rewarded pro productive and useful work is narcissism, then you need to either learn about work or narcissism.
I like the sentiment. Keeps a lot of people out of politics so the few can rob everyone else blind. No no, you don’t want this job it’s for loser hacks only.
Ah the irony, I'm blocked from viewing that page by Cloudflare
Performing security verification
This website uses a security service to protect against malicious bots. This page is displayed while the website verifies you are not a bot.
Incompatible browser extension or network configuration
Gmail has a system of reputation as you suggest. It is very likely that enough people marked their emails as spam, which the OOP could figure out on the postmaster dashboard if they were so inclined: https://postmaster.google.com/managedomains
It also goes the other way, if enough people click "not spam" and interact with your mails, your reputation gets better. I'm currently trying to do that with my personal email/domain - will probably take some time though. For now, my friends say my mails land in spam even though I get a 10/10 score on mail-tester.com and similar sites.
When I ran my own mail server, I was lucky to even make it to the gmail spam folder. More often it didn't even make it that far. From what I can tell, O365 is even worse though.
They should go through, at least to spam - but your setup needs to be flawless, meaning you need to correctly set up the 'holy trinity' SPF, DKIM, and DMARC.
Sites like mail-tester.com, learndmarc.com, or sending a mail to ping@tools.mxtoolbox.com (which will reply a report to you) are pretty useful for that.
But yeah I have only limited experience I suppose. Having some mail correspondence with friends in the hopes of improving my domain's reputation to those mail servers.
Oh and btw, I relay through my cloud providers mail delivery system - doing it from your own IP is probably a whole different league.
I run my own email server on own ip since 2005. Never had any issues with gmail or M$. Changed IP at least one time, no difference from the first day on.
Just one time I had to activate SSL for inter server communication. But that was a known thing that gmail was rejecting you otherwise.
s/average/median
reply