gexla

Recent commit...

LICENSE | Relicense under the Apache License 2.0 | 7 minutes ago

Honestly wonder if creating a website for something like "Jason of 404 media won't help you with tech support requests" would help this issue. Probably worth whatever the effort might be involved. But you can probably seed it around also. Drop it in some Reddit threads as like a sig, etc.

You changed your post. You mentioned Levine. I don't know what else you changed. But Levine posted this in his article. I don't know where you're getting the part where he can sell half his shares.

"If Zuckerberg needs to pay taxes, he can sell his new zero-vote shares to raise money without diluting his voting rights. Not too many: He can “immediately sell or transfer up to 20% of the shares of Class C common stock that he receives in the Dividend, representing 10% of his overall economic stake in the Company, for any reason,” but the board “negotiated for and obtained Mr. Zuckerberg’s agreement that 80% of the shares of Class C common stock that he receives in the Dividend be subject to the Stapling Condition,” under which he can’t sell those no-vote shares without also selling some high-vote shares. The basic deal remains intact — he needs to own a meaningful chunk of shares to keep voting control — but there’s a liquidity buffer for taxes."

Where did you get this information? We know how this works. This is a public company. This all happens in the open.

https://www.sec.gov/Archives/edgar/data/1326801/000162828026...

We don't have to post hyperbole or speculation (there's social media for that.) This is a public company, we can read public documents such as...

https://www.sec.gov/Archives/edgar/data/1326801/000162828026...

NOTE: Look for "Founder Alignment Agreement" in the above doc as an example if you're doing a quick scan.

https://www.sec.gov/Archives/edgar/data/1326801/000162828026...

https://investor.atmeta.com/investor-news/press-release-deta...

Is this hyperbole? Or did you actually look into details? I'm curious about your understanding of what happened here. I believe this is an exchange of share types, but I could be mistaken.

I guess you didn't see the video of a car sitting on a ramp upside down, doing flips in the air, jumping over a semi and other cars...

Odd world we live in when we don't see that as being cool. But, we're seemingly headed there. At some point, it's going to be hard to imagine something that could surprise me.

Nice, you caught my eyeballs at the right time. I'm going to sign up for it. Cursor / GrokBot already gives me this, but I'm looking to recreate this setup as a backup so that I don't get locked in. This feels like the right direction. Move all this stuff off my desktop and then run everything from my phone.

Okay, so you felt like someone was telling you what words to use. I didn't feel that.

I don't get why using “super intelligence” is an issue here. It seems the US admin sort of politicized this and now US companies have their own reasons for selecting terms that may not make sense to people who aren't making the same decisions.

Everything else seems reasonable? Assume the worst and put the controls in place to handle that scenario? Isn't this how we approach security normally?

NOTE: NOT a security expert, so I may be talking out of my...

Right, easy comparison to make the the open source community for software.

And it's not like this is something where we're loaned some top math genius for a limited amount of time and we have to make the most of it. Rather, this is a new high water mark. The accessibility of the results is no longer scarce. The scarcity has shifted, and that's where the focus of the math ecosystem should shift as well. And it doesn't help for a frontier community to saturate and take over messaging pipelines that were typically managed by the math ecosystem. It's not about "stay in your lane" but rather "we need coherence and be careful not to break the system."

Just two cents from someone who could screw up basic cashier math on any given day.

This also gives you some surface area for LLMs to help you out in writing rather than code. Have it give an adversarial response or whatever you think is important. Then you can also see what is obviously wrong with the LLM's argument along with some nuance or interesting points you may have missed. Then you can have it pull in upstream sources of knowledge done on the topic, which is something that has been game changing revealing at times for me. "Here's the text from which all else flows, it was written 60 years ago, notice that nothing on this list has been written in the last 20 years." You know you got a keeper when that happens.

This may be partially why the US dispatched the LA Rams and the SF 49ers. Even though McDonald's has hundreds of US bases in Australia, the corp lacks a rushing and passing game. I don't see the NFL being successful there, but time will tell. Interesting note, in SE Asia, I watch NFL commentators based in the UK. It's strange seeing NFL talk with a British accent. And TIL. I always thought the Hungry Jack's was a Burger King knockoff, but apparently it's actually Burger King with a different name.

Let me grab my calculator and add up the time I have spent reading about model releases since Fable has been released. It seems they all place themselves relative to Fable. I'm sure that time has added up to far greater than 10 seconds. At some point, it ceased to be a meaningful differentiation. This is especially true when I put the model through real usage.

Seems they should be making it more clear they are not associated with Herdr. Just as I use a minimal harness (Pi) I appreciate the minimal features of Herdr. Much of the features this adds, I prefer to bring my own. In many cases, I let my agents go and don't much look at Herdr.

The minimal decisions is partly what gains my trust. If Herdr were to come up with something like this, then I would give it more attention. Since it is not associated with Herdr, I'm no more likely to give it attention than any other the other overwhelming AI coding things that hit Github in too short of a period.

If everything basically rivals Fable, then why is everything still using it for comparison?

Interesting how the author compares the situation to smoking. Were any of the execs of the tobacco companies asked to step down because smoking was found to be harmful to health? My guess is they simply continued and adapted based on how they could remain a profitable company with continued shareholder satisfaction.

I don't see how it makes sense to use models outside of Grok or Composer in cursor unless you like the tooling to such degree that you're willing to pay out the nose for access to OpenAI and Anthropic to use with that tooling. And Codex is regularly scoring towards the top among harnesses, so you might as well use the best harness available for the given model. I'm happy with Grok and Composer in Cursor. If Cursor wants to add more models, they should host more open models.

GPT gets the most out there, but creates interesting nuance that turn an idea into a more creative exercise. Claude is like a more grounded GPT but may miss the nuance. Gemini is my last pick, but does better than the rest for making something clear and understandable. It gets to be exhausting parsing through Claude outputs and breaking it down into something more easily understandable. One way I use to improve this is to ask the LLM to pull upstream ideas from prior work on the topics. This makes me feel better about the possibility of hallucination, gives me alternate places to look, but the model still may pull things out of context or fall over on the interpretation.

Given all that, I can definitely see how Gemini would be preferred. And good for Google, because I would rather my offering be the top choice for the most people rather than better serving a small subset of users.

My understanding was that it was totally an experiment to create something that works like Cursor's Origin or whatever it's called. Just the idea of it is probably not something you would care about for your own usage.

I don't know about real successful. Since you mentioned Langchain, you could look at https://www.langchain.com/dcode which is a CLI harness build off Langchain deep agents.

Yeah, whatever it is, it's particularly bad at front-end from what I have been seeing.

I remember back in the day when you could apply this to Google search results. When I was freelancing, I had a number of people I worked with who would fire off fast questions, but they would still pay me for my time. Then I would return what was essentially a Google search result shortly after. Yes, I would send my own answer, not some paste from Google.

But the point is. They had search, so why didn't they think to do the search? They had the same tools as I did. And yet regularly they couldn't think of the answer. I still deal with this with AI today. When you ask a question, maybe you're asking for the interpretation of the question by the person on the other side. And it's that interpretation that goes to the LLM. Or otherwise some filter of how that person thinks about the question differently than you do.

Just skimmed the article, but the CF your employer used was probably around a lot longer than AI? There's parts of it that can bring everything down. Then there's parts of it that will bring down just what your team(s) put in a load of hard work to build on top of it.

I have found in my own experimentation, I create way more trash than I did before. If I'm 10X creating trash, then I'm WAY less productive creating the stuff that I actually use.

I found a funny thing before. I'm in the 2100 block of Github IDs, meaning I was OLD SCHOOL. This got me thinking about how I was probably one of the first users of the first GPT model to be in a major product, Copilot. According to Grok, Copilot was GPT 3. I figured it as earlier because it SUCKED at generation other than auto-complete.

I'm now thinking the best way is to use these models to create the scaffolding and keep my brain in the architecture with strict reviews and small PRs. Slower, but less slop. Basically, just using AI for a bump or two above what I used Copilot for back in the day. Less running agents all day creating slop that I'll never look at. More with serious focus on what I bring my full attention to. Increased productivity, less BS.

Maybe that's just rearranging chairs on the deck on the Titanic. But it's what I'm thinking. And hitting send! ;) That's probably not the win, but I think that path could reveal it.

This is super confusing. It's not supposed to have a Linux version and it's supposed to be only for Supergrok Heavy (during beta according to the docs) and yet the landing page gave me an option to download it for Linux and to setup a trial account that I could use it with via Cursor. But, it doesn't seem to get through the gates to connect to whatever it needs to connect to.

Otherwise, I think this would still go into my "I don't get it" folder. It seems like the flavor of the day is "agents can talk to each other" and Grok Bot is selling that point along with Claude Code. But I think Pi has had this for a while? Or it was easy to wire it in.

The page design looks like it has Claude's fingerprints on it. Claude design is the new bootstrap.

Here's another article that explains the guy called in an expert who told him "at least $100K crashed through your roof."

Doesn't mention if he actually sold it though, unless I missed it.

https://www.nytimes.com/2026/07/15/science/meteorite-new-jer...

KeePassXC also has one.

I don't think we're far enough into this game to know what winning even looks like. If the US frontier labs are losing, it may have been their own missteps on handling such massive change of being the fastest growing apps ever. Until China can ever show me something new rather than just cheaper, then I just see efforts from that region as a force of commodification. OpenAI will go down in history as the company who brought what many understand to be AI to the first billion people.