Chris Olson

LGA tower frequency is 118.7. JFK has two tower frequencies, 119.1 and 123.9. I don't think this was a turn of the dial mistake; I think the pilot used muscle memory and just hit the wrong muscle.

Tesla is putting on quite the show in downtown Austin with the announcement of Terafab.

Operational AI has to move across systems, maintain context, and decide what happens next. @mach1ai powers complex real-world processes with automation that doesn’t just run. It thinks.

During the holidays, Instagram should offer a feature where you can scroll nothing but ads for gift inspiration.

@mshamsho @eightsleep That’s horrible. I probably won’t get one anymore.

I'm a lifelong Democrat. But Gavin Newsom’s recent tactics remind me of a marketing quote: when you start copying your competitor, you end up looking like their asshole.

@GrantM Maybe YouTube hasn’t been optimized for playback being at 10x ;)

Why does the YouTube app make my phone heat up so much? It’s been happening for months. Same videos played on Spotify are fine. It’s so odd that such a popular app would have this problem. I assume it’s my hardware, but it’s not that old.

Isn’t this just like when FB rolled out a redesign? History shows it makes noise, but never stops more people from using it. That’s a pretty good place for OpenAI to be.

GPT-5 in voice mode describes my custom instructions before it actually starts answering my questions. “Absolutely, let’s dive into that with a bit of that Ben Thompson meets Adam Liptak meets Heather Cox Richardson kind of vibe.” Strange.

@benhylak I always add "Reduce the number of adjectives you use when responding" to my custom instructions. I don't know when it exactly it happened, but all of these models have become quite verbose...

so far, i find the output of grok 4 to be long, winding and a little incoherent. there are often really good bits and chunks. maybe there's something about the writing style not sticking for me. it sounds really good, but often not that useful?

@elonmusk it’s ready for a wider net!

MCPs are reinforcing that prompt engineering will be around for a while, right? Or maybe they can only be reliably used with good reasoning models.

The Essential Role of Operators in AI Adoption The dawn of AI agents has prompted discussions about their role in the future of work. While some imagine autonomous systems handling complex tasks independently, practical deployment shows a different reality: businesses need skilled operators to oversee AI agents. The greatest businesses deploying AI capabilities will be those that adopt technology that helps them effectively oversee and manage their AI to achieve success, rather than simply releasing autonomous agents with minimal oversight. The goal is transforming businesses through AI, moving beyond merely artificial intelligence toward artificial capabilities—where AI is already smart, but operators must enable it to be truly capable. Understanding how AI agents work in business reveals they need attention. Working with these agents should feel like managing a person. The operator becomes the point of contact between your business and AI models, translating business needs into AI actions. Rather than autonomous behavior, it's a sophisticated form of delegation and instruction, demanding human intelligence to formulate and refine. A collaborative cycle emerges where AI can recognize its shortcomings, but humans must interpret these signals and adjust the AI's parameters. The power of this model lies in its economics: one skilled operator can oversee AI agents that accomplish what previously required entire teams, creating efficiency gains that transform businesses. This points to a clear conclusion: the future of AI in business is not one of manager-less operations, but rather a dynamic partnership where skilled operators guide AI to achieve unprecedented scale. This approach may not be what everyone believes, but I'm convinced it's the path that will succeed long term. The market will ultimately decide who's right, but my experience shows that businesses investing in technology for their operators to oversee, manage, and continuously improve their AI capabilities will be the true winners in the AI era. The future belongs to those who embrace this symbiotic relationship between operators and AI systems.

I got the opportunity to share my experience deploying AI agents across Trace's operations, transforming processes that paved the way for @mach1ai's foundation. Thanks for the great conversation @jjacobs22! youtu.be/kVcn9acgpuU?si…

@marccampbell Yesterday!

When did Grok 3 become available in the API. I can't believe I missed this.

@jonfavs I know this is what you were getting at with the reference to the 8th, but just to really hammer the point home: it feels like either way you look at it, it’s both cruel and/or unusual…

Yeah, so here's why the Constitution doesn't allow the state to round up Americans and send them to be tortured in a foreign gulag because they lit cars on fire: 1) The 14th amendment means the state can't strip you of citizenship - whether naturalized (the case you cite) or U.S. born (post-14th, Congress did not have the power to strip citizenship from anyone born here). Deportation is, by definition, for non-citizens. 2) The 6th amendment means the state can't imprison you without due process, which at the very least includes a trial (courts have ruled this also covers non-citizens) 3) The 4th amendment means the state can't just bust down your door without a warrant because they suspect you may have a bad tattoo 4) The 8th amendment means the state can't send you to a brutal dictator's prison where inmates have been tortured and killed just because you vandalized a car made by the president's top advisor Also, grow the fuck up. You're 48 and you talk like a 5-year-old who just learned his first bad word

@Bolt__AI What is the secret behind the document analysis for o1? Is it using OCR to process the document and putting it into context?

BoltAI v1.23.0 released ✨ Some notable changes: - The o1 series models now support Document Analysis - Estimate token & cost for Gemini models - Reworked the chat input field: faster and better - Other bug fixes Update when you can ✌️

@daniel_nguyenx @Bolt__AI @AnthropicAI Does it auto cache, or is there something we need to do in the bolt interface to activate this feature?

Claude Prompt Caching is amazing 🤯 When sending a subsequent message: • without cache: uses 35k input tokens, costs ~$0.1 • with cache: uses just 497 tokens, costs ~$0.01 ( cost for cache reads: 35k x 0.3 / 1M = $0.01 ) Available in @Bolt__AI v1.18.3 ✌️

@_heydave @NotionHQ From Ivan’s tweet about upcoming launches. Surely this says Notion Mail?

Just found something interesting in my Finder after the latest @NotionHQ update 👀 Anyone else?

Is there a way to reset the X "algo"? It has become nothing but TikTok/Reel videos. I know some might say, "well, it's because that's what you're engaging in." But it's really not; I feel like if I watch one video or get sent one in a DM, it's all I see.