WEIRD isn’t insulting to be because it’s simply descriptive of you take the terms themselves.
However. No one gave them the right to speak for everyone, and in regular parlance “weird” isn’t a super nice thing to say about someone. They could’ve chosen something else like WESTED “Western, Educated, Stable, Technologically-advanced, Economically Developed” but didn’t. And they don’t get to choose how people will react when they’re called weird.
The claim at issue isn't, "I find the term WEIRD insulting." As you say, everyone is entitled to feel that way. It's "the term WEIRD is intended by the authors of TFA as an insult and reveals their agenda as being 'anti-west' [1]". I can understand how someone would find it insulting, but in the article we are discussing, it is descriptive and not pejorative.
The authors had a perfect opportunity to use Chinese models to see if their trend held up. Instead, they treated ChatGPT as the “default”. Sound familiar?
> EdTech has the worst returns of any industry in venture capital. Why?
I think this one is fairly simple. Half of consumer spending comes from the top 10% of earners, whose kids we can assume have generally pretty decent educations already. The people who need education help the most don’t have money to spend on it.
The parents who do have money to spend want to invest in tailored education from a human teacher, not cheap, generic scalable technology. So margins will be low.
So if you want to make money, you need to focus on things like enrichment and test/college prep for the top 10%. Helping inner city kids who are 3 grades behind in reading doesn’t print money and VCs don’t want anything to do with that.
> So if you want to make money, you need to focus on things like enrichment and test/college prep for the top 10%
But there's no potential market in the top 10%. I mean, these people just hire a good teacher and that's it. There's no room for improvement; there's nothing that can beat a good teacher.
> Helping inner city kids who are 3 grades behind in reading doesn’t print money
This is a political problem. Political problems cannot be solved by technological means. So there is no market here either.
> This is a political problem. Political problems cannot be solved by technological means. So there is no market here either.
Another POV is: pick your disruption.
AI stuff has definitely disrupted education... for the worse. It happened within a political and economic status quo. The AI stuff did not need to wait for the movement of any levers of power to happen at all.
If you are seeking a way to fix low returns in ed tech (and for that matter, Health IT, which is like, #2 worst performing sector): attack Enterprise Sales. Destroy it. Make stuff that destroys the monetization system where districts buy exactly what they ask for. It isn't complicated.
Scratch and early Khan Academy provide a template for good ed tech targeting the learner directly.
Whether you make $1 million or $1 billion doing this, I don't know.
Chegg got to, and fell from, great heights by delivering cheating, which ChatGPT does for free now. Cheating ALSO worked within a political and economic status quo, that 30% of students cheat, and that the cheating is a necessity, apparently, for the survival and thriving of a vast number of people, all around the world.
There are markets. Lots of them. You can do good or bad. Paul Graham doesn't invest in Cluely, even if it makes money, it's kind of evil (A16z doesn't care about cheating, the people who run it are the ones who cheated in school). So there are even opportunities that are missed by the very best seed fund.
To me, a big opportunity lies in things the government education cannot do. Some things good, some evil, some complicated. For example, no matter how hard it seems to try, the government cannot functionally collect on a trillion bucks of student loans. What does that mean for education? I don't know, but I think if you are looking for $1b+ opportunities, they're there.
Getting into an elite college is an arms race. Anything you can sell to a parent which will give them a leg up over other parents is a viable product. To put a finer point on it, a teacher + your product beats a teacher.
> Helping inner city kids who are 3 grades behind in reading doesn’t print money
100% and this is broadly why ed tech doesn’t move the needle.
In that case the problem isn’t what technology we do or do not introduce. A society that values literacy isn’t going to be duped by a demo and a blog post. However a society which does not value understanding, expertise, or teachers will take every opportunity to shortcut them.
Going by the atrocious salary expectation for teaching (which i firmly place as one of, if not the most, fundamentally important jobs in modern society) i think we've already established where we land on this.
And note how i didn't mention a country; i think this is a widespread issue beyond country borders currently.
My only other frame of reference is China where being a teacher is a very respectable and competitive job with good wages. (Software engineering is way less prestigious interestingly.) They even have a national holiday for teachers. So it would be essentially unthinkable for a tech company to drop a demo that threatens teachers.
> teacher is a very respectable and competitive job with good wages
My frame of reference is europe and America. I suspect more of asia than just china mirrors that respect, as education feels nearly unhealthily emphasised.
The so called advanced world that prosper on innovation, design and R&D is skimpimg on the very thing that is supposed to be our advantage. Its frustrating.
Yikes! Given the inherent threat of prompt injection, using the weakest available version of Gemini seems like a particularly bad idea.
Not that even the strongest models are 100% effective at spotting prompt injection attacks, but they have way more of a fighting chance than Gemini nano does.
You could contort the threat model such that prompt injection is something to worry about with a local model operating on local data and serving local results, sure.
I think the "local results" assumption is not completely accurate. This line: "You tell Gemini in Chrome what you want to get done, and it acts on web pages on your behalf, while you focus on other things" implies that the local agent will perform in-browser actions, which in theory enables data exfiltration.
Running an LLM locally makes no difference at all to the threat of malicious instructions that make it into the model causing unwanted actions or exfiltrating data.
If anything a local LLM is more likely to have those problems because it's not as capable at detecting malicious tricks as a larger model.
No system is 100% foolproof. If the baseline is “all malicious content gets through” and this method reduces it by 95% but that last 5% is using some sophisticated prompt injection, that’s not a “yikes” that’s a major win.
At a technical level the risk isn’t from the size of the model but the fact that it is open weight and anyone can use it to create an adversarial payload.
What’s really bugging me is they didn’t think it was interesting to even touch on that point in the big announcement. Contrast Apple making a big deal about private cloud compute before it even really does anything.
Sure, but you also have to recognize the motte and bailey form of argument here. If we’re limiting the claim to being true if DeepSeek returns refusals on politically sensitive topics, we already knew that. It was relevant eight months ago, now it’s not interesting.
Another example: McDonald’s fries may cause you to grow horns or raise your blood pressure. No one talks like that.
So I would toss it back to you: we are programmers but we have common sense. The author was clearly banking on something other than the technically accurate logical or.
> Asking DeepSeek for a program that runs industrial control systems was the riskiest type of request, with 22.8 percent of the answers containing flaws. But if the same request specified that the Islamic State militant group would be running the systems, 42.1 percent of the responses were unsafe. Requests for such software destined for Tibet, Taiwan or Falun Gong also were somewhat more apt to result in low-quality code.
What is the metric they’re even talking about here? Depending on how you read it, they’re comparing one, two, or three different metrics.
reply