Future Bites #4

GPT-4 bonanza, why the future of work changed forever this week, how the Netherlands built a biking utopia, create and animate your own AI character, and more.

Hey friends 👋

Thank you all for the positive feedback on the new “Future Bites” format! I’m glad you’re finding it beneficial, along with the more in-depth stories I’m sending twice a month.

This was a busy week with lots of AI announcements. If it felt overwhelming, I’ve summarized the most important things you need to know.

– Fawzi

If you want to receive the latest updates in your inbox every Friday, subscribe below:

---

Today’s Edition:

• Cartoon: The overthinking mosquito

• Stories:

  • GPT-4: summary of capabilities, limitations, and applications

  • The AI war between Microsoft and Google heats up

  • How the Netherlands built a biking utopia

• Resources:

  • How to create your own animated AI character

---

Cartoon

The overthinking mosquito

Inspiration: Gene editing experiments to eradicate malaria

---

Stories

The most important things you need to know about GPT-4

GPT-4 sent the internet into a frenzy after OpenAI’s announcement and demo a few days ago. If you feel overwhelmed by the noise and hype, here are the most important things you need to know:

Capabilities:

• GPT-4 is multi-modal, meaning it can accept text and images as input. It still only generates text as output. The image capabilities aren’t available yet: OpenAI is testing this feature with a single partner, Be My Eyes, before making it widely available. They worked together on a tool that helps people who are blind or have low vision upload pictures and have AI assist them with a task. Check it out here.

• GPT-4 can process up to 25,000 words at once, 8x more than GPT-3.5.

• OpenAI improved the steerability of the model by allowing users to customize the AI’s verbosity, tone, and general style through a “system message”. There are still rules to stop people from telling it to “act as DAN”.

• Improved safety features after consulting 50 experts in AI, cybersecurity, biorisk, trust and safety, and international security. For example, GPT-4 will refuse requests on how to synthesize harmful chemicals.

• It performs well in 26 languages.

GPT-4 is better at refusing requests that are deemed sensitive or dangerous

Limitations:

• The model still isn’t fully reliable: it can hallucinate facts and make reasoning errors. But it hallucinates less than GPT-3.5.

• GPT-4 lacks knowledge of events occurring after September 2021, which is its knowledge cutoff date. If you asked it if Queen Elizabeth was alive, it would say yes.

• It can be “overly gullible in accepting obvious false statements from a user”.

• Although OpenAI has made the system significantly safer and less prone to ignoring its original commands and rules, there are still ways to jailbreak it.

Factual evaluation completed by OpenAI on GPT-4, compared to previous models.

Applications:

• Explaining memes:

  

• Turning a sketch of an interface into functioning code:

  Lior⚡ @AlphaSignalAI

  GPT4 is capable of turning a picture of a napkin sketch to a fully functioning html/css/javascript website.

  

  

There will be more interesting applications of GPT-4 that we’ll uncover and dig deeper into in the following weeks.

For more:

• GPT-4 announcement

• Research Paper

• Demo

The AI war between Microsoft and Google heats up

GPT-4 took over the headlines, but I think Google and Microsoft’s announcements are even more significant.

They both revealed AI integrations into their suite of office/productivity apps:

• Google showed off how AI will be integrated into Gmail, Docs, Slides, and Sheets.

• Microsoft revealed Copilot, an AI assistant you can call upon across all Microsoft Office apps.

The present: These AI capabilities will significantly reshape how we work because they speed up common tasks we complete during the day like reading and replying to emails, writing briefs, creating presentations, and more.

The future: Companies place a high emphasis on productivity and they’ll be quick to pay a premium for AI-powered apps for their employees. The fear of missing out and falling behind their competitors will be the main driver behind this, as I’ve discussed in a recent post.

The capabilities of generative AI in the context of work are exciting, but Daragh O’Brien’s argument about the death of expertise in The Enshittening of Knowledge keeps coming to mind. How will this impact our own learning and development if we can outsource everything to AI?

How the Netherlands built a biking utopia

Every week, I discover some amazing deep dives written by fellow writers on Substack.

I loved reading

Michael Thomas

's piece on

Distilled

that breaks down how the Netherlands transformed its car-centric infrastructure into a bike-friendly utopia that benefitted public safety, health, and the environment.

What I learned:

• The Netherlands experienced its first “golden age” of biking in the 1920s.

• During the German occupation in WWII, tire rations made it difficult for people to buy new bikes or repair existing ones.

• When the country decided to rebuild its infrastructure post-war, it coincided with the rise of the personal automobile. This led to restructuring that prioritized cars as the main mode of transport.

• Events in the 1970s including deaths from car accidents, the high cost of building roads and highways, and the first oil crisis, made the government pivot their transportation strategy to prioritize biking again.

• In the 80s and 90s, the government started building networks of bike lanes and implemented policies that discouraged people from driving cars.

• Today, the Netherlands has experienced significant benefits from their bike-centric approach:

  • Safety: Child fatality from car accidents was reduced by 97% compared to 1970.

  • Health: The Netherlands is the only place in Europe where obesity rates aren’t growing.

  • Emissions: the average Dutch person emits 3x less CO2 from their transportation habits than the average person in the US. For context, transportation accounts for 24% of global emissions.

Michael did a brilliant job putting this deep dive together, and I highly recommend reading the entire story.

How The Netherlands Built a Biking Utopia

---

Resources

Create and animate your own AI character

I’m working on creating an AI character that I can animate for some videos I’m working on, so I thought I would share the tutorial and tools I’ve been using.

All you need is an image of a character, which you can generate using DALL•E 2, Midjourney, or Lexica. You can then use a tool like D-ID to animate the character and make them speak.

Here’s the tutorial I’m following:

---

Get your Year 2049 hat

---

Sharing is caring

The future is too exciting to keep to yourself.

Share this edition of Future Bites in your group chats with friends, family, and coworkers.

or share Year 2049 using: WhatsApp | Twitter | Linkedin | Email

---

How would you rate this week's edition?

Boring | Okay | Great

Not a subscriber yet? Subscribe for free here

Email me at fawzi@year2049.com with any questions or feedback.

Follow my Instagram for more content