I’m happy for the Cursor team, but I sincerely don’t understand the valuation.
Most people I know who figured out how to use Claude Code or Codex directly get much more compute and a much better power-user experience. The difference is not even funny. And it’s not an IDE vs terminal because both claude and codex work just fine in vscode.
What annoyed me most was my own experience: last November I paid $20 for Cursor, burned through the quota in one day after just a few prompts, and didn’t find any clear way to even see quota, so i contacted support. Somehow they claimed I had used around $70 of compute and implied I should be grateful. But they calc price of all input tokens as if they were cashmiss tokens, which is obviously more expensive than real API usage and this was extremely dishonest on their part.
So from my perspective, Cursor or similar solutions often look like a middleman between the user and model provider. In theory, that should be hard to defend against using the frontier labs directly.
That is why it is mystery for me why so many people still pay for it over orders of magnitude cheaper claude/codex
Revenge! A friend of mine was scammed this way two years ago.
So when I finally received a similar offer on LinkedIn, I made them pay to me, twice XD
They sent me a GitHub repository and claimed they needed help deploying a token or something similar. The obvious goal was to get me to run their code while connected to a wallet with real funds.
I told them: sure, I can probably help, but my time costs money.
So I convinced them to pay me for a consultation. For half an hour, I explained how “amazing and simple” the process was, and how they could do it themselves. They claimed to be from the US, but on the call they had an extremely heavy Asian accent. I even recorded part of it on camera, because the whole situation was ridiculous.
Before they could get too furious, I told them I did not have more time and that they would need to book another call. So they paid me a second time.
Before the second call, i got myself absolutely new macbook was for fun testing it, basically a fresh Mac (i should have been using vm because it’s faster to restore but that time i wasn’t familiar with UTM and i got parallels which is scam itself), just to see how their scam worked.
The scam was classic npm install-style: get someone to run scripts from a Git repo and hope they are unaware of how much access that can give to their machine.
Honestly, two years ago, I also did not fully understand that simple ‘nom install’ could give attackers such deep access to your computer.
basically this - it took the social media industry quite a few years to invent a simple-looking thing, short-form video with discorver algo and it became extremely addictive for a huge part of the population.
Now imagine an AI that knows your personality type, good at psychology, always supportive. Best friend ever. That creates a far stronger moat than any price or technical advantage of an AI coding tool.
what would be really cool (maybeme i missed it?), is for your tool show clear useful info on the "top panel", not just logo. Like mac-stats or littleSnitch do.
Could you please provide any proof for: “if you look at any large company’s CEO, they’re there because they took these hits or provided plausible deniability for jig boss in the past”.
Never thought from this angle, interesting theory
I’m curious about the reasoning behind this approach. Surely, Elon Musk and his team are smart enough to recognize the legitimacy of such research. Initially, I thought this kind of content was just playing to the lowest common denominator—a form of low-effort clickbait. However, this strategy risks alienating more educated or critical followers.
Given Musk’s intelligence and the capabilities of his team, I’m struggling to understand the logic here. Is it meant to drive engagement by polarizing opinions, or is there some deeper strategic intent that I’m missing?
I'm pretty sure you could find justification even for why top models don't support microSD card expansion. The problem with this line of argument is that if they wanted to, they could support both without any issues. The real reason is money. It's more profitable to have Bluetooth only when you also make AirPods, and not include storage expansion when you sell built-in memory options at a 400% markup.
*typo: you can send more than one bit per photon on average
I'm very curious to learn more about 1cm, what is the math behind it? Do you speak about classical music CD with ±700mb of capacity? I was always fascinating by ability of old super scratched optical disks still functioning without problems.
that's truly incredible how much lives can be saved with good electronics and it-systems.
I wish the war ends asap, before the autonomous ai-powered drones comes into play... because at that point this might become complete skynet-like hell.
A little offtop for everyone who uses comparisons that "old good days" computers/whatever were more expensive so new generation don't have moral ground to complain. Please don't forget about overall context. In that old day you could afford to buy house and don't spend the most of the income for renting. So it's incorrect to simply compare prices for one specific niche, overall context for expenses should be counted in.
a counter take from me, as someone who uses adblocks everywhere possible but used to work for facebook ads department:
ads aren't inherently bad, irregardless of much data do they track. What really bad is how this ads is targeted and how much ads is delivered. I would LOVE to see relevant to my interested and situation ads. I like fpv drones and often see interested ads on instagram and led to a purchase. The problem after that I can't just edit the vector of my interests for ads targeting and keep bombarded with something not relevant to my situation anymore, and boy so much of it I see.
I really don't know why big ads network didn't move forward with user controlled ads mechanics - because we all humans buy something anyway, it's okay of being aware of the options on the market. It's not okay to mentally torture users with irrelevant, annoying and overwhelming amount of ads.
what does it mean in simple words? as one of many StackOverflow contributors (although very small) I don't want that my answers were wall-guarded by SO website, they are free to use advertising revenue from traffic generated by my content to remunerate for creation and supporting the platform, but the content itself is mine, why it shouldn't be that way?
The most unthought response I see every time. “But what you expect cop gonna do”? There are millions things they could do if given insensitive. Imagine a startup that get paid for every busted criminal? Put a honey pot, install cameras, track license plates. It can be resolved in matters of weeks.
Most people I know who figured out how to use Claude Code or Codex directly get much more compute and a much better power-user experience. The difference is not even funny. And it’s not an IDE vs terminal because both claude and codex work just fine in vscode.
What annoyed me most was my own experience: last November I paid $20 for Cursor, burned through the quota in one day after just a few prompts, and didn’t find any clear way to even see quota, so i contacted support. Somehow they claimed I had used around $70 of compute and implied I should be grateful. But they calc price of all input tokens as if they were cashmiss tokens, which is obviously more expensive than real API usage and this was extremely dishonest on their part.
So from my perspective, Cursor or similar solutions often look like a middleman between the user and model provider. In theory, that should be hard to defend against using the frontier labs directly.
That is why it is mystery for me why so many people still pay for it over orders of magnitude cheaper claude/codex