Mozilla.ai did what? When silliness goes dangerous
tante.cc
tante.cc
Sussman attains enlightenment
In the days when Sussman was a novice, Minsky once came to him as he sat hacking at the PDP-6.
“What are you doing?”, asked Minsky.
“I am training a randomly wired neural net to play Tic-Tac-Toe” Sussman replied.
“Why is the net wired randomly?”, asked Minsky.
“I do not want it to have any preconceptions of how to play”, Sussman said.
Minsky then shut his eyes.
“Why do you close your eyes?”, Sussman asked his teacher.
“So that the room will be empty.”
At that moment, Sussman was enlightened.
Training a randomly wired net is like pretending that the training data won't give it pre-conceptions, when the training data is what drives the model.
(Pretending that closing ones eyes, removes all items from existence)
Sussman has written a whole program to play Tic-Tac-Toe: two players, a 3x3 grid, players alternate turns, the game ends under such-and-such conditions, etc. etc.; and then he thinks that by such a superficial gesture as wiring the net randomly he somehow creates a tabula rasa with no "preconceptions of how to play." Minsky points out, by an equally superficial gesture, that giving up control over one minor aspect of the program (the wiring, the ability to see with one's eyes) really has nothing to do with the issue at hand (how much "understanding" of Tic-Tac-Toe has been baked into the program, how many things exist in the room).
Or, to put it back into koan form:
One day, Sussman wrote a program that had no preconceptions of how to play Tic-Tac-Toe. The following day, Sussman wrote a program that had no preconceptions of how to play chess. His two programs were not the same.
That said, I don't see any difference between dot-com bubbles busting, regardless of decade, and them not busting. From my perspective, I see stupid(along with good) all the time.
> On March 20, 2000, Barron's featured a cover article titled "Burning Up; Warning: Internet companies are running out of cash—fast", which predicted the imminent bankruptcy of many Internet companies.[48] This led many people to rethink their investments. That same day, MicroStrategy announced a revenue restatement due to aggressive accounting practices. Its stock price, which had risen from $7 per share to as high as $333 per share in a year, fell $140 per share, or 62%, in a day.[49] The next day, the Federal Reserve raised interest rates, leading to an inverted yield curve, although stocks rallied temporarily.[50]
Keep the kool-aid flowing to avoid it, this can't happen twice, wait, what was that in 2008 again?
I’d be annoyed too. At least this guys angry blog post is somewhat educational - reminding us of biases and how NOT to use LLM’s.
all this being said, time to go outside and touch some grass. Get off the computer for a bit.
> "On two occasions I have been asked, 'Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?' I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question"
Like, people have known this doesn't work since literally the First Industrial Revolution. And yet still we try asking the magic oracle to unbias our data.
If you read the blog post it's pretty obvious what happened. Someone at Mozilla.ai had an extra day on their hands and ran a bunch of text they'd collected through a few models. They thought "hey, this is kind of cool, let's make a blog post about it". Then they wrote one stupid line about their motivations (likely made up to justify playing around with local models) and get completely lambasted for that one stupid line.
I'd rather live in a world where people are comfortable throwing together a quick blog post detailing a fun/stupid project they did than one in which they do that anyway but are hesitant to share because people will rake them over the coals for being "unserious".
to be fair, it's worth lambasting them over it because they are perpetuating the myth that AI is bias free (which a lot of people actually do believe!) and putting the weight of mozilla's reputation behind it
Remember when ProPublica claimed that recidivism prediction models were biased against black people? (https://www.propublica.org/article/how-we-analyzed-the-compa...)
They rigged their analysis. There are two different fairness metrics - individual and group-based - that cannot possibly be fulfilled at the same time if there is any difference between the groups. (https://arxiv.org/abs/1609.05807v2)
The model was fair all along according to individual metrics. ProPublica picked their fairness metric so as to make the model appear biased. The whole thing was a lie.
The same is true regarding humans though, so "use a magic ai to avoid bias from humans" still doesn't make any sense.
That's certainly a possibility, but the post sure looks like it's trying to claim otherwise.
Man, modzilla really do the most useless things. I'm really surprised just how bad they're at generating any profit, the silly ideas and products are wild.
Can't they just try and make Firefox the best, ubiquitous and one day content actually against Chrome?
But that's not what I read the Mozilla post that this post is ragging on as saying.
https://hachyderm.io/@inthehands/112006855076082650
> You might be surprised to learn that I actually think LLMs have the potential to be not only fun but genuinely useful. “Show me some bullshit that would be typical in this context” can be a genuinely helpful question to have answered, in code and in natural language — for brainstorming, for seeing common conventions in an unfamiliar context, for having something crappy to react to.
> Alas, that does not remotely resemble how people are pitching this technology.
Yes, people get bamboozled because LLMs are trained to bamboozle them, Raskin didn't call them "a zero day vulnerability for the operating system of humanity" for nothing -- but that's all there is.
If this is true, then I’m a court jester, because none of my projects started as serious work by a serious organization. And ML wasn’t lame until everyone started taking it so seriously.
The key with ML is to have fun. Even the most serious researcher has this motivation, even if they won’t admit it. If you could somehow scan their brain and look, you’d see that all the layers of seriousness are built around the core drive to have fun. Winning a dota 2 tournament was serious work, but I’ll wager any sum of money they picked dota because it seemed like a fun challenge.
If the author is looking for a serious AI organization, they should start one. Otherwise they’re not really qualified to say whether the work is bad. I have no opinion on Mozilla’s project here, but at a glance it looks well-presented with an interesting hypothesis. All of my work started with those same objectives, and it’s mistaken to discourage it.
The more people doing ML, the better. It’s not up to us to say what someone should or shouldn’t work on. It’s their own damn decision, and people can decide for themselves whether the work is worth supporting. Personally, I think summarizing a corporation’s knowledge is one of the more interesting unsolved problems, and this seems like a step towards it. Any step towards an interesting objective is de facto good.
Bias has become such an overrated concern. Yes, it matters. No, it’s not the number one most important problem to solve. I say this as someone raising a daughter. The key is to make interesting things while giving some thought ahead of time on how to make it more inclusive. Then pay close attention when you discover that some group of users doesn’t like it, and why. Then think of ways to fix it, and decide whether the cost is low enough.
There is always a cost. Choosing to focus on bias means that you’re not focusing on building new things. It’s a cost I try not to shy away from. But the author seems to feel that it’s the single most important priority, rather than, say, getting a useful summary of 16,ooo words. I think I’ll agree to disagree.
Ruins a company’s credibility…
If you thought that any researcher has a clue what they’re doing, I’m afraid you’ve been mistaken. We have hypotheses and observations from past experiments, but no one has any idea whether something will work until they try it. So by discouraging them from trying, you’re decreasing the likelihood that any useful work will be done at all.
Is it groundbreaking? No. But the author's overwrought political rant about Mozilla, AI, the internet, and probably capitalism seems unwarranted based on a small blog post. From the "about" page of tante.cc seems like they are some kind tech/political/leftist/"luddite" commentator.