Adam

@elonmusk I struggle with Grok 4 Fast writing code that can look write, but is wrong. For example physics calculations a lot of times will not take into account time, but instead be based on frames. So it looks like it works, but its flawed.

What are the most important improvements to make to Grok 4 Fast? Critical feedback is much appreciated.

Reading the comments it seems like people believe they are talking to Elizabeth Holmes. Most accounts like this are managed by outside teams. They may correspond with what should be posted, but its highly unlikely she has access to X.

If you want to read more check it out here: raleon.io/blog/dtc-ai-be…

the one that surprised me the most was Claude 3.5, it just isn't good at these kind of tasks, where 3.7 is surprisingly good:

Introducing a new AI benchmark to evaluate AI models for Direct To Consumer specific tasks. If you have any models you think we should evaluate please drop them below so we can add them to our benchmark runs.

@klaviyo is there an issue with your api's, status's show green, but across multiple stores we are seeing this issue.

There is an unknown connection issue between Cloudflare and the origin web server. As a result, the web page can not be displayed.

This is the API with an issue. a.klaviyo.com/api/profile-bu…

It is possible badly tuned models are being hosted, but seems only somewhat likely as even meta ai website has poor coding. Maybe it works well for other things other than coding?

Llama 4 is a big disappointment for coding currently. I hope this improves, but seeing it top LMArena for coding something fishy is going on. The model was either highly over fit to LMArena questions or LMArena is being manipulated. It’s not even in the top 20 in my tests, and doesn’t work at all in things like RooCode.

@lcfritz 100% agree you should my post I just made about them! I have a roof leak they keep running me around in circles...

Here’s a consumer tip - avoid @BlueRavenSolar at ALL costs-they are so deceptive in their business practices & the info they share. My solar panels haven’t been working & the trouble shooting isn’t working - they want $350 to come look at it. BIGGEST MISTAKE OF MY LIFE!!!

@BlueRavenSolar I really need someone to escalate my roof leak. I reported it March 11th, and was told i'd hear back on the 12th. I didn't so I called and they told me I'd hear back on the 13th, I didn't. I called again today and just keep getting the run around. Now Ill hear next week, not even a day, NEXT FREAKING WEEK???? I hope you realize how much more you are going to need to repair if more damage keeps happening. You are incredibly fortunate I've mitigated a lot of the damage. This is after an incredibly painful experience. 1. You messed up submitting paper work to my energy company. Didn't even tell me, I had to call you weeks after waiting only to be told oh we messed up we'll get it on it now. Then Duke responded with a failure because of insurance paperwork. Which I figured you'd deal with, and didn't so I waited days, and then called. 2. Then the insurance paperwork wasn't right, when I called I got a condescending sounding person telling me that Duke Energy is picky and needs specific wording. Why wouldn't you know that before submitting and wasting a month of time? I was told it would be resubmitted but someone was out of the office. I waited days and had to call back again to get it moving. This is after emailing it like you said, only to be ignored with no response for days, so I had to call again to get any movement! 3. Now, i'm told that I'll hear back NEXT week about a roof leak, WTF. That isn't even a repair or someone coming out here to look at it, its you'll hear back next week. To say I'm annoyed is a massive understatement, considering we had installation done for months now, still aren't energized due to your companies mess ups, and now have a leak that you won't even rush to get fixed. On top of this you don't even communicate back to me, I always have to call. I'm probably the most patient person you could deal with and even my patience is gone at this point. If anyone is considering solar from you, i'd be very skeptical we were even lied to by the salesperson about tax deductions (don't get me started on that) and have been treated incredibly poorly by your support. You have a chance to do right by this, and i'll give you the benefit of the doubt, but holy crap I can't believe i'm dealing with this.

@QwQiao Dang i'm sorry for the company impacted :(

@brian_armstrong Same!

The key attribute I look for in hiring for these companies is "builders" who thrive in ambiguity. Can they prototype solutions in days, not weeks, and then refine them with a sense of urgency.

15 rules of vibe coding with Cursor

My personal challenge is that in crypto i really only see 1 true PMF, and that is gambling. Maybe there are active users that I haven't stumbled across yet, but almost everything that is or has done well, has some aspect of speculation/gambling. I think the market has mostly spoken. But what i'll add is, a lot of people forget about building traditional businesses with crypto rails. It can make a ton of sense to not build crypto first but instead take a traditional problem and build a solution that is better because of crypto. The problem with that is founders see people launching coins and getting rich with very little effort, while they are working 80 hours a week trying to build something real. I've always felt that people that take the easy path or cheat end up losing in the long run, but it doesn't really seem to be the case fully in crypto yet.

Trying to decide if crypto is still the right place for me: 1. I believe crypto is a great financial rail for expanding financial freedom, but that's mostly true for places without existing infra. (e.g. underdeveloped countries went straight to mobile payments, skipping cards) 2. For places with mature financial infra, most of the value for crypto is in speculation. Arguably there is value to bringing 24/7 Vegas to everyone, and probably positive sum too because of entertainment value. 3. I have lost conviction over time on crypto being a decentralized compute platform to work on this for 5 years+, nor do I believe that consumer behaviors desire this more than the alternatives. Based on the above, I don't have the passion to build financial rails in developing countries and I don't have long term conviction in crypto being a compute platform, so that leaves speculation-based applications. I don't honestly have anything against speculation, but I am just not a degen at heart. I tried getting into it many times, but it was very difficult to hook myself to it. And in this case, it's hard for me to imagine winning in an increasingly crowded market (literally hundreds of pump clones) when I can't sympathize with my potential users.

I know the time we are in right now is short lived, but we have so many cheap/free AI tools right now that we are going to look back and miss these days in the future. Much like the days of $5 Uber rides.

@burkov I feel you!!!

Note to myself: 24 months aren't enough for the planet to understand that a machine learning model can only generate information it was trained on.

@seatedro I'd be curious what you think of my testing methods. youtu.be/uFzpI3UO0uM I found Google Pro 2.0 Exp to be really good, it starts off kinda weak but does amazing at the end.

tired of seeing people drop new LLMs and say it's great at coding and the demo is just simple webslop so I wrote my own benchmark, zero-shot 10 tests 1 Typescript 1 Python 3 Rust 2 Zig 3 C Gemini 2.0 Pro is the best right now, with 4 passing tests. GPT-4o is the worst.

My daughter's @Roblox account was compromised and had 8,500 Robux stolen. While thankful she got 1/2 what she saved up back, its also disappointing that Roblox doesn't have better measures in place to punish people that clearly stole from her. Its also kind of crazy to me that they won't fully give her back what was taken. At least she got part of her presents back, but it does make me and wife consider not buying her anymore as there is very little protection it seems. Just a warning out there for anyone.