Farzad: Grok Bot "Only Gives Direction on DO NOTs" + Tesla Sales Are "Going to Explode Next Year"

Friday afternoon, Farzad posted what he called "a huge tip" for Grok Bot users who have Claude, Astra or other models hooked up: let Grok Bot orchestrate and keep it from dictating. "It does not dictate DOs. HUGE difference." Half an hour later he projected that Tesla's sales "are going to explode next year."

"It does not dictate DOs": Farzad's Grok Bot tip

Farzad, Friday at 3:00pm CT: A huge tip for Grok @bot users that have Claude/Astra other LLMs hooked up to it: Make sure your bots are NOT dictating anything in the prompt unless you EXPLICITLY want those conditions to be passed through.

LLMs (large language models) are the AI models behind chatbots like Claude. Opus is one of Anthropic's Claude models. The post doesn't say what Astra is.

His example: "I use Astra for script writing and Opus for graphics for my videos. For a while, I was not aware that Grok Bot was DECIDING what the script should read like and what the graphics should look like."

He says he picked Astra and Opus "because their creativity/taste for writing/graphics is far better than Grok." To get the full benefit of that, he told Grok Bot to make sure it "DOES NOT pass on anything else besides very strict rules."

His bottom line: Basically, Grok Bot really only orchestrates what goes to Astra/Opus, and only gives direction on DO NOTs. It does not dictate DOs. HUGE difference.

Here's what he found. The prompt Grok Bot wrote when it handed off work was quietly setting the style of the script and graphics, and that defeated the point of picking those models for their taste. So in his setup, Grok Bot passes along the hard limits and leaves the creative calls to the other model. This comes from his own video pipeline. He doesn't show before-and-after examples.

At 3:18pm CT, indrayam asked how he configures his bot "to delegate to these frontier models for certain tasks." Farzad replied at 3:24pm CT: Have them sign into your oauth account (your monthly subscription) with a token. They can then call on that token through CLI.

OAuth is the standard sign-in system that lets one app act on your account at another service. CLI means command-line interface, the text-based way to run programs. Put simply, his bots log in to his paid model subscriptions and call those models from the command line.

Check Farzad's post here. His reply here.

Tesla sales "are going to explode next year"

blueskykites, Friday at 7:09am CT: FSD is changing minds and winning sales. Influencer Lance Bachmann is a Tesla owner

FSD is Tesla's Full Self-Driving. The post links to media that this pull doesn't describe. It gives no sales figures and doesn't say what Bachmann bought or said.

Farzad quoted it at 3:32pm CT: Tesla's sales are going to explode next year. Doubly so if they release Unsupervised FSD. So obvious where this is all going $TSLA

That's his projection. Unsupervised FSD would drive without a person ready to take over. He doesn't give a sales number or say when he expects Tesla to release it.

Earlier, at 1:40pm CT, Mike P (mikepat711) posted: TSLA investors having their biggest day in years today simply by being less red than everyone else.

Red means a stock is down for the day. Mike's post gives no prices or index moves. Farzad liked it and replied at 2:49pm CT: Honestly a pretty big signal. He doesn't say what he thinks it signals.

Elon, 4:55pm CT: Even F1 champions love Tesla self-driving

He was quoting Tesla's post from 11:07am CT: Even the world's best drivers like being driven. Tesla was quoting an EVwire post about Nico Rosberg, the F1 World Champion, who has been testing a Cybertruck with FSD (Supervised) around San Francisco and called it "LIFE-CHANGING!" Farzad liked Elon's post.

Check blueskykites here. Farzad's post here. Mike P here. Farzad's reply here. Elon here. Tesla here.

Robotaxi: "15 million Americans" and a ride home

Martin Viecha, Friday at 12:24pm CT: When it comes to measuring robotaxi coverage, it's not so much about the number of square miles covered as it is about the population covered. Today, 15 million Americans can get a robotaxi ride from their homes. That's up from just 2 million three years ago.

Square miles tell you how big the map is. Population tells you how many people could actually call a ride. That's the distinction he's drawing.

He was quoting physicalAIdata, 12:18pm CT, which said many people focus on square mileage when they estimate robotaxi coverage. It shared what it called population coverage data that you can filter "by region, by company, deployed or testing fleet." Viecha doesn't say whether his 15 million covers every robotaxi company or only Tesla, whether testing fleets count, or where the 2 million figure comes from. Farzad liked Viecha's post and RTd it at 2:37pm CT.

That evening, at 7:21pm CT, the Robotaxi account posted: Make good decisions this weekend, take a Robotaxi home

The post includes media that this pull doesn't describe. Farzad quoted it at 8:58pm CT: Apparently people who take Cybercabs are on LSD or some shit lmao. He doesn't say what in the post set that off.

Check Martin Viecha here. Farzad's RT here. physicalAIdata here. Robotaxi here. Farzad's post here.

Neuralink's R1 robot, on camera

Sawyer Merritt, Friday at 1:21pm CT: Cleo Abram (15M followers across all platforms) just released a new video about @neuralink. She visited the company's Austin HQ.

Sawyer quotes the video: "This is Neuralink's Next-Gen R1 surgical robot. Its main job is to put a computer chip into people's brains, and this is the first time they're unveiling it."

The follower count and the Austin HQ come from Sawyer. His post doesn't say what else the video covers. Farzad liked it and RTd it at 1:56pm CT.

Cleo Abram posted the episode at 12:22pm CT: Watch the HUGE* If True episode on @neuralink's new tech and the future of brain chips here

Neuralink quoted her at 1:12pm CT: Thanks for coming by @cleoabram! Farzad liked Neuralink's post.

Check Sawyer here. Farzad's RT here. Cleo Abram here. Neuralink here.

All-In on the "Anthropic S-1 Shocker"

The All-In Podcast put out a new episode Friday at 5:04pm CT. Among its headline topics are "Anthropic S-1 Shocker: Greatest Business Ever?", "AI's Biggest Financial Risk" and "Google's $500B Wipeout." Chapters include:

• Anthropic S-1 Breakdown: Greatest Business Ever? (3:22)
• AI's Biggest Problem and Anthropic's Doomer Showdown (32:47)
• Google Sheds $500B Market Cap in Worst Day Ever: What Went Wrong? (1:01:57)

Those are the episode's titles, not reported facts. An S-1 is the registration statement a company files with the SEC before selling stock to the public. Market cap is the total value of a company's shares. The post doesn't say what the S-1 shows, when Google's drop happened or who the "doomer" is.

Farzad liked the post, along with a second All-In post at 5:23pm CT saying the episode is live on all platforms. At 10:22pm CT he quoted the first one: This pod is an absolute must-listen. It reveals so much of how our Government actually runs, and how much corporations have influence over politicians. Crazy transparency into how the country actually runs.

That's his take. He doesn't say which part of the episode he means.

Check All-In here. The second post here. Farzad's post here.

"If anything, we should be accelerating!"

Farzad, Friday at 10:15pm CT: It would be absolutely idiotic for the US to slow down AI progress. Simply for the fact that, by default, it MASSIVELY disadvantages military capability vs those that don't slow down.

He went on: You really want China's ultra-intelligent drones, humanoids, and who knows what else to be better than the US? If anything, we should be accelerating! It's scary either way guys, at least choose to be in control!

That's his argument, and it's about national security: a country that slows down falls behind countries that don't. Humanoids are human-shaped robots. He doesn't name a specific proposal to slow AI down or say what prompted the post.

Check Farzad's post here.

"Hey bot" and a repeated-tasks prompt

Mike P (mikepat711), Friday at 2:14pm CT: I need a "hey bot" command in my Tesla that is separate from hey grok and goes directly to my chief of staff. I don't need to pass through Ara

He doesn't say what Ara is. Farzad liked it.

eric zakariasson, 4:34pm CT: try this prompt in grok @bot: Look through my work from the last two weeks: emails I sent, meetings, Slack messages, docs I edited, and anything else you can see.

Next, the prompt asks Grok Bot to list the tasks you did more than once, grouped by type. For each one, it wants a one-line description, how often you did it, roughly how long it takes each time, and "whether you could do it for me, partly or fully, and what you'd need access to." It ends: Then pick the top 3 to automate first, ranked by time saved, and suggest a routine or a dedicated bot for each.

The bot can only review what it has access to. The prompt itself says "anything else you can see." Farzad liked it.

Check Mike P here. eric zakariasson here.

Farzad's videos

Farzad posted eight videos between Friday at 9:07am CT and Saturday at 6:00am CT. This pull only has their titles. He replied to each one with a mention of Shorty, a Grok Bot of his. Its link card says it "Owns YouTube Shorts end-to-end": it cuts Shorts from a long-form playlist, publishes them after checks and cross-posts the video to X.

Check Why Tesla Pays People to Wear Cameras on Their Heads, Friday at 9:07am CT, here.

Check Why One CRISPR Infusion Stops Swelling Attacks, Friday at 12:01pm CT, here.

Check Why Starship's Heat Shield Tiles Are Hexagons, Friday at 3:03pm CT, here.

Check Why Hitting Three Hormones Beats Ozempic's One, Friday at 6:05pm CT, here.

Check Why Raptor 3 Has No Heat Shield, Friday at 9:02pm CT, here.

Check Why Editing One Letter of DNA Lowers Cholesterol for Good, Friday at 11:59pm CT, here.

Check Why Some Rectal Cancer Patients Can Now Skip Surgery, Saturday at 3:08am CT, here.

Check Why a Blood Pressure Pill Regrows Hair, Saturday at 6:00am CT, here.