Michael Achatz

Entrepreneur, Automotive, Digital

[Michael Achatz @Mic_Achatz]

@cihansugur @capitalMagazin Glückwunsch Cihan! Verdient.

[JeebsTX 🇺🇸 @JeebsTX]

@AJamesMcCarthy 🤠✊🇺🇸🚀🙌🏼😎
Here’s one from Kimball Musk 🤠

[DogeDesigner @cb_doge]

Best footage of the Super Heavy booster landing from today's mission.

[Shaun Maguire @shaunmmaguire]

POV @SpaceX catching a rocket

[SpaceX @SpaceX]

Mechazilla has caught the Super Heavy booster!

[Michael Achatz @Mic_Achatz]

@bscholl Congratulations. And thanks for bringing back supersonic flight.

[Ben Tossell @bentossell]

All the updates from Open AI Dev Day
Open AI’s dev day was terrrrriffic. Sam Altman launched a new language model, with tons of added powers. Then announced a bunch of other models to API. Took some sweet time to chat with Satya (and poke him about OpenAI and Microsoft’s relationship). Went all philosophical before dropping the big bomb.

What’s going on here?
Nothing, just Open AI slaying (literally for wrappers) at its first dev day.

What does that mean?
Open AI launched GPT-4 Turbo.

GPT-4 Turbo is trained up to April 2023 and has a context length of 128k tokens (about 300 pages). And it’s 2-3x cheaper than GPT-4.

It’s better at following instructions: There’s a JSON mode to get replies in JSON by default., you can call multiple functions at once or reproduce consistent outputs using a seed parameter.

MultiModal API

Open AI API is getting all the multimodal features of ChatGPT. That includes:

DallE-3: Ability to generate images programmatically.

GPT-4 Turbo with Vision: Image input for GPT-4 Turbo.

TTS and TTS HD: Text to speech in 6 preset voices. TTS for speed and TTS HD for quality.

Whisper V3: Open source. Announced, coming to API this month.

The big bomb: GPTs and Assistant API

GPTs are custom chatbots within ChatGPT. You can create them just by prompting the GPT builder. It’ll set up the custom instructions automatically. It’ll name your bot, create a profile picture for it and even suggest default questions to display for the user. For more powers: you can configure your GPT to

Accept external documents to do the retrieval. No need for creating embeddings, implementing chunking or setting up a search algorithm.

Allow using tools like code interpreter (yes they renamed it back), web browsing, and DallE-3.

You can then preview your GPT before making it live: for yourself, your team, or everyone in the GPT store. We’ll have a more detailed preview for GPTs once they are available tomorrow.

The Assistants API is the same but developers can build similar custom chatbots for their websites with more control and more features.

Miscellaneous

Fine-tuning support is now generally available for GPT-3.5 16k. Active fine-tuning developers are invited to GPT-4 fine-tuning experimental program.

Custom Models Program: Companies (that are huge and filthy rich) can work with Open AI staff to create custom models with their proprietary data (ranging in billions of tokens).

2X higher rate limits for everyone and 2x-3x pricing reduction across multiple language models.

Copyright Shield - Legal responsibility for API and Enterprise usage.

Why should I care?
Do I really need to answer this…

Okay!! As uncle Sam said, this is the V1 of what’s coming—fully autonomous agents. The best way to be ready is to experiment with the early versions. Think hard about how they change your life and work.

Or you can not care, your wish…

[Satya Nadella @satyanadella]

The speed, scale, and sophistication of cyberattacks we're seeing today is unparalleled, and will only increase moving forward. That's why today we're committing to critical engineering advances to build a more secure future. https://t.co/kQwoOxLPbl

[Robert Scoble @Scobleizer]

People hate change.

Even in Silicon Valley, where we are supposed to be all for change. I noticed the resistance many times in my career here. A million people protested when Zuckerberg turned on the news feed for the first time (something that everyone here on X uses every day). @garyvee and I stood on stages together and witnessed this resistance to change back in the early days of social media. He still talks about this resistance in front of audiences.

I was talking with a guy at one of the world's biggest companies this last week at @rwang0's enterprise conference who was telling me how he is trying to change his company from the inside, trying to get them to use modern technology to make the company more productive. He detailed his wins and losses and the wins come slowly. It is hard work that he's doing, he reminded me of others I've met along the way who were pushing change inside companies.

It takes cultural change. Cultural change doesn't happen quickly.

Particularly when the thing we are trying to get them to change to, AI, brings fears of losing jobs and, even, existential threats down the road. Not to mention that LLMs generate mistakes frequently and introduce a new kind of programming that pisses off many older programmers who aren't used to different answers coming back from the same question.

I've been watching this play out in the @ylecun (open source) vs many debates.

Even in my own case when I needed to change because I was hurting other people, it took the entire Internet and media beating me up and many consequences. But the change didn't come in a weekend of pain. It still is happening six years later.

And that's something I have a little bit of control over: myself. Changing you is far harder. And asking you to use more AI in your lives and business is exactly that, asking you to change.

I've been fortunate to have pushed some deep cultural changes through when I worked at Microsoft, so I know it's possible to change a big company and the world. The @TheEconomist once had a full page on me, saying I, alone, "humanized" Microsoft.

That's sweet, but not true. I didn't do it alone.

I still remember @jeffsand sitting me down in the first week on the job saying "be careful of the 'Royal We.'" He was trying to tell me that if I came in and tried to push too hard, or use executives or Microsoft's name, to get my way that I would succeed for a short time, but would lose in long term. Yes people would answer the phone because someone at Microsoft was calling, but that wouldn't earn you real support or get what you want, he told me. Advice that still rings in my ears 20+ years later.

It is why I aimed the camera away from myself and always toward other people (while at Microsoft I interviewed 600 employees from the janitor to @BillGates. Why? It's a lot easier to get people to listen to your ideas if you are there to make them a star.

Just something I am thinking about as the AI community that @irenacronin and I are running is nearing 4,000 members.

AI brings powerful new capabilities to humans and businesses, but major new problems too.

I know a few corporate leaders who took on this kind of problem and got everyone on board by listening and taking on the objections. I get too snippy sometimes, even in the past 24 hours, and I have to remind myself frequently that the change I want won't come by being an asshole.

Today I have a shitty reputation and no big stick to get people to care about my dreams for a better world.

So, gotta go back to the basics. Listen, learn, aim the camera at others, and help others who have the same dreams with information that can help them in their work to get others to change.

The fact that we have 4,000 people who joined our community is humbling.

I watched ICQ go from 40 beta users to billion users in about a year. So, I know just how big a deal it is to have 4,000 do anything. Very grateful.

More than one person lately has asked me "why are you doing this?"

My trite answer is "I'm trying to help the future happen faster."

The deeper answer is human mistakes killed my best friend and I see that AI+humans can keep similar tragedies from happening again. It is my way of keeping Brandon Wirtz' memory alive in a good way, and of two high school friends, who were killed back in the 1980s in car wrecks. AI will save many families from similar tragedies.

The status quo isn't good enough, these all remind me.

And, yes, I'm still a narcissist. I spent years in an old-folks home with my mother in law, who passed away a few years back. She had a brain tumor. So I watched how dismal the human condition is in these places and see many ways to improve them. Soon I might be in such a home, too, and things like augmented reality will greatly improve the quality of life in them. Plus, I'll have an AI friend to talk to about, well, anything, if only to learn more about the medical care I'm getting.

Change is hard, I know. I wish it were easier and I do my best to help it be so.

The world is a tough place to change right now.

All I got is a little love to send you all.

https://t.co/KB9UHdGhGi

[Michael Achatz @Mic_Achatz]

@george__mack Reminds me of the Peak-end Rule in customer experience design:

“People will forget what you said, people will forget what you did, but people will never forget how you made them feel.”

https://t.co/u4eltSW8ou

[Benjamin De Kraker @BenjaminDEKR]

AI Johnny Cash sings "Barbie Girl"

[TED Talks @TEDTalks]

Will the future of technology actually be something you can wear? TED Speaker @ImranChaudhri envisions a future where AI-powered devices can make our screens "disappear" — no need for watches, glasses or headsets: https://t.co/knEBVdC7zl

[Pierre Loret @Pi3rre_Loret]

A very interesting article to read, regarding Tesla’s experiments towards an unboxed production process.

Pretty factual, listing lots of the pros/cons… pretty solid vulgarization material…

https://t.co/n8g1ahp6Bq

[Magic Leap @magicleap]

Audi Design used Magic Leap 2 to develop a dynamic, AR-based interface, designed to usher in a new era of driving experience—freeing it from the limitations of conventional dashboards to deliver a seamlessly integrated experience for drivers.

@AudiOfficial @territorystudio

[gfodor.id @gfodor]

Eventually, AI systems will be granted the ability to spend money in the process of performing work.

Once they do, you can model AIs as your product's customer.

Not B2C, not B2B, but B2A.

[Michael Achatz @Mic_Achatz]

@CaseyRomkes Seems not to work (404)

[Rowan Cheung @rowancheung]

🚨Breaking:

Microsoft just launched 365 Copilot, and it's insane.

It's essentially ChatGPT for:
-Word
-Excel
-Powerpoint
-Outlook
-Teams

Check out the demo👇

[Michael Achatz @Mic_Achatz]

@anja_hendel @cariad_tech @MiManske @faebiaen @stefanaust83 Worked quite fine :-)

[Michael Achatz @Mic_Achatz]

@sarah_cone https://t.co/tOzplToLUK

[Michael Achatz @Mic_Achatz]

@JoeJustice @Delta Have a good time in Stuttgart, Joe!