alt Hacker NewsI think most users of websites like reddit, x, and yes even HN don't realize how much traffic is inorganic. Marketing firms, government agencies, and many other interested parties with money to burn are absolutely aware that you search "best {product} reddit"
I've commented on this before, but I strongly suspect much of the narrative around AI is being formed with strong inputs from these patterns. What's your basis for thinking that codex is best for planning, but opus is best for implementing? Is it based on extensive experimentation and first hand experience in a non-deterministic environment, or is it that you saw a large number of people on HN and X say that?
Why was the dominant narrative on cursor coming within spitting distance of opus with a MUCH smaller team and less capital "LOL THEY USED KIMI!!" instead of "wow, open source models + a bit of RLHF training and some clever context management got within spitting distance of the industry giant and way cheaper"? The latter sentiment is a whole lot more damaging for a company eyeing an IPO with existing investors with very deep pockets.
Replies
A good reason to find specific individuals with relevant knowledge and follow their writing directly.
Think simonw and his pelicans... but there are lesser known trustworthy voices as well. It just takes some time to find them for a given area of interest.
Also bring back blogrolls.
> Why was the dominant narrative on cursor coming within spitting distance of opus with a MUCH smaller team and less capital "LOL THEY USED KIMI!!" instead of "wow, open source models + a bit of RLHF training and some clever context management got within spitting distance of the industry giant and way cheaper"? The latter sentiment is a whole lot more damaging for a company eyeing an IPO with existing investors with very deep pockets.
This comment is interesting because you took a narrative that was being pushed and marketed (Cursor was close to Opus) and accepted it as the ground truth.
The dominant narrative I saw around that, at least in my bubbles, was disappointment when they actually tried it and discovered it was not, in fact, close to Opus.
> Why was the dominant narrative on cursor coming within spitting distance of opus with a MUCH smaller team and less capital
And how do we know that? How do we know Cursor is "withing spitting distance of opus" (whatever it means)?
Let me guess:
> that you saw a large number of people on HN and X say that
this is why oldschool chat > social media
curating for trust and expertise and diversity of opinion
related: Cursor composer line of models is so good relative to cost. "auto" served me just fine until they recommended Composer and I've been continually happy with it. Then Claude Code with Opus dropped and everyone went bananas and I gotta say I just assumed I'm too casual to know how bad Cursor has been?
But then I think maybe not really? Granted, I'm not orchestrating 100 Agents doing overnight work. But relating this to your point, if the CC-camp + HN hadn't proclaimed otherwise, I would have no idea what breakthrough CC+Opus made. (Cursor was first with plan mode right?)
> I think most users of websites like reddit, x, and yes even HN don't realize how much traffic is inorganic.
Came here to say this - I have always been extremely cautious and assumed most things online were just marketing tactics. But I never realize how far and how strategic some of these campaigns are.
I’ve recently started really getting my hands dirty with marketing for an app I’m building and the things I’ve learned in the past year have made me questions many of my views on things. At some point you realize that it’s all marketing or some form of effort to exert influence.
A good book somewhat related to this is Attention Merchants
Remember Quibi?
All the money in the world can't actually turn a turd into a market leader.
If you have a good product you have to play the marketing game to avoid getting left behind. If you have a bad product you try to play it and you still don't get picked up. (This last bit is where things usually turn into an argument about "no, obviously [this thing I don't like] is bad and is only popular because of the marketing", which assumes taste is more universal than it is.)
> What's your basis for thinking that codex is best for planning, but opus is best for implementing?
I for one work on an agentic product where we use all 3 of the major frontier models. The models absolutely have preferences and "personality" that lead to different characteristics.
In my eyes:
* Gemini - consistently the best at pure reasoning and tunability. Flash models are particularly good at latency sensitive small-scale reasoning. The tradeoff is they struggle with some basic behavior, like tool calling.
* Claude - consistently good at long standing sessions. Opus may or may not be the best model, but it was the first model that crossed the "holy shit" threshold. I understand it's quirks/nuances and it's consistently solid. It's the best for me because I've learn how to be incredibly effective with it.
* ChatGPT - Probably really good, but probably not worth switching from Claude. Last time I used their frontier model, it was a bit random. It would have moments of brilliance immediately followed by falling flat on it's face.
>I've commented on this before, but I strongly suspect much of the narrative around AI is being formed with strong inputs from these patterns.
"The AI talks down to me like Reddit because it's trained on Reddit" has been a running joke/quip/gripe on the "less refined" parts of the internet for awhile now.
At a previous company our marketing team had a $50k/mo budget with an agency that got their basically verbatim posts posted by all the tech blogs like TechCrunch, venture beat, Huffington Post etc. I got really aware of the tech media and I read every story as intentional marketing.