Brad Hutchings on LinkedIn: Tesla's Beer-Serving Optimus Robot Was Controlled By A Human The Whole Time (2025)

#MeWriting Driving doesn't require a lot of thought. Elon's genius is his focus on solving problems with these technologies that the technologies can solve.All the things I'm reading, he is getting a lot of heat for keeping it simple and sticking with cameras. He seems mostly happy with raising the miles between interventions metric, as deaths, injuries, and accidents per mile are reduced commensurately. Everyone of us would be safer overall giving control to FSD in the new vehicles today, just not in the particular situations where we would individually be safer. We'd also be safer with more FSD vehicles around us, especially as they displace distracted and impaired drivers.He's taking the exact opposite approach of the LLM crowd, which can't find value in what the systems actually do, and tacks on layer upon layer of ineffective and expensive garbage pretending those deliver that value. His cars and robots can't count the Rs in strawberries either, but he won't promise that they will.I've read so many people disappointed with the robots. The complaint is that they get the humanoid form and motion right, but they don't do much. Elon talked about them doing "tasks", hinting that those would be the "apps" for these systems. Conventional developers and artists should be champing at the bit for the upcoming opportunity!Elon takes the easy wins. He simply zigs when everyone else has jumped ahead to zagging. He takes what the technology gives, and usually no more. When he overpromises, it's on timelines of the doable. It's what frustrates everyone who gets frustrated with him. But it's also how he's established his track record of eventually winning with pretty much everything he touches.I'm zigging on Elon. I know how unpopular that approach will be today, but it's the easy win.

17

32 Comments

#MeWriting LinkedIn peops, your assignment for Friday morning is to splain how AI can be so good at driving, yet at the same time suck at counting Rs in strawberry. #YouGotSomeSplainingToDoI look forward to your posts.

3

5 Comments

#MeWriting Work would be a lot easier if this were true. It's not. ☠️ If you have people in charge who are not leaders, you will get a toxic virtual office.🦨 If you have people in charge who don't value people, you will get a toxic virtual office.🪓 If you have two traditional standup meetings per day, you will get a toxic virtual office. Same if anyone wants to meet with you after 3:30pm in the afternoon. Same if people regularly end up on-call or doing work outside 8-4 in their local time. Same if people get upset about long lunches or picking up kids or walking dogs or general day to day distractions.⚰️ If, by some weird cosmic chance, you're on a customer call including the boss and you get a text that a very good friend of yours passed away and the boss doesn't get you off that call in under ten seconds, you are in a toxic virtual office. I will never get over that one.👊 Remote virtual companies are difficult. They must be high trust. They must be high performance. They need to be balanced with frequent in-person opportunities. A good culture takes effort. It cannot be an afterthought.

5

#MeWriting Side note: If you automate stuff, you should have a musician on your team. Musicians know to ask, "What could possibly go wrong?" Because it does go wrong.

5

#MeWriting I'm not worried about AI girlfriends. I'm worried about AI boyfriends. After 10 minutes of thinking about it, I'm pretty sure they could become the basis of the next wave of feminism. Like today. No improvements to the models needed. Guys, if you haven't found her yet, you are screwed. There is no chance of winning. And if you have, well, the bar on you just got raised to impossible. They haven't even built AI boyfriend yet. I probably shouldn't have posted this. Some tech bro just read it and started a slide deck.

4

2 Comments

#MeWriting I'm working through a short list of general themes about how to approach AI. Here's one that's a rough sculpture today.AI (based on LLMs) will help you achieve 2nd percentile crappiness quickly and without much effort. We need to recognize these accomplishments with something along the lines of "that's adorable" and "bless your silicon heart" and encourage them to learn more. A budding programmer who has only ever been a startup CTO should be encouraged to explore what actually goes into developing software. Same with budding Picassos who dabble in DALL-E. Let's show them how layers work in Photoshop.I do pity the users who are sucked into believing they are expert enough using AI. But more than that, my heart aches for the developers who enable instant crappiness. Not that I haven't done that my whole career. But there were always a few rank amateurs who could and would find the springboard and launch themselves to greatness.Sculpt with me in the comments.

12

10 Comments

#MeWriting People really want this AI stuff to work. Here's the reality though. If I need 100,000 top 2% intelligent people in 25 years, the best strategy is still to get 5,000,000 couples to conceive tonight.Believing in people is the ultimate e/acc. 💪

9

5 Comments

#MeWriting Twitter exchange with someone who was part of something very notable, but is not terribly famous himself... I don't want to light him up. I want to point to something most are missing with LLMs.He used GTP o1 to gather reasons on both sides for participating on X (nee Twitter). Then he pasted both arguments in and asked it to decide which was more compelling. I suggested he do it 10 times, categorizing both the yes/no decision and the reasoning given. He said he was familiar with the limitations of LLMs. And a lightbulb went off in my head. 💡🤕At surface level, I came off as critical of his letting an LLM make a decision for him based on one response. Sure, I don't want a world where serious people are doing that. Or even unserious people like me. "You accelerate at random, Pinky." Please world, don't make my favorite young adulthood cartoon come to life like that! 🌎 What I've found in approaching decision problems with LLMs is that you can use their "a plausible answer" replies to get groupings of plausible answers, worded and emphasized a little differently, that evaluated together, show better answers. There's nothing mystical about it. It's expected behavior from the algorithm! So if I'm trying to make a decision, those groupings become a better basis for decision than any single generated plausible answer.These are the kinds of breakthroughs I help my clients with. Connect and DM if you're interested.P.S. A fun post would be to compare the algorithm in the picture below to how inference works. ⚾🏏 <-- How is there no baseball bat, and I have to use cricket emoji?!? That was a home run aside if you missed it. Please re-read.

1

#MeWriting A problem with anthropomorphizing LLMs is that it can lead you, as a self-identified leader and strategerist, to come up with dumb ideas that mess with people's lifestyles. #SeeAuthorsBylineThe #hashtag #IsNotDead. It is used as a language feature far more than as a connector in message databases. See the first hashtag in all of my posts. It has worked its way into spoken language as well. Hashtag #HowTheEffDidYouMissThatStrategyGuy?!?! <-- Read that out loud. And, so ironically in this context, it has worked its way into LLMs! See posts on LinkedIn obviously written with ChatGPT that have clever hashtags that have nothing to do with linking to other posts.But I'm most annoyed with this post by one of our AI lifestyle strategy leaders because it's only possible if he decided that: "We are now at a stage where machines can properly interpret the meaning of a message. They don’t need guides or training to do this anymore."Wrong. They do not properly interpret meaning. Look at the data structure behind LLMs: the parameters and weights. Here's the funny thing. We generally don't know what any particular parameter in the neural net "means", other than the ones at the edge which are tokens, just pieces of words.If you look in his comments, I asked for a clarification before writing this, and I got a flippant "read the news" response from our dear AI leader. I also re-read his post after my second coffee this morning to make sure it wasn't a clever sendup. Because a clever sendup it would be with all the supporting comments! #PerformanceArt #NopeItsNotP.S. Hashtags were popularized on Twitter. About a decade ago, a friend asked me over dinner to explain what a hashtag was. That's when I knew they weren't a database search feature. When did you figure that out?P.P.S. If you or your business need help staying safe from horrible ideas put forth by AI leaders and strategists, connect and DM me. #ThisIsntCuteAnymore

1

1 Comment