Brazil at Silicon Valley Takeaways

TL;DR OF THE TAKEAWAYS (wow, you really don’t have time to waste, right?)

AI Affecting Jobs and Skills: AI is poised to transform text-heavy and mechanical jobs, requiring new skills and frequent re-skilling. It won't create a mass extinction of jobs, but the tools people use at work will change and everyone needs to adapt.

Use Cases and Adoption of AI: AI is increasingly adopted in corporate settings for improving productivity, but Brazilian companies will incorporate AI much slower then the US, 48% vs 87% of firms have AI as their priority, respectively.

Future of AI: Change and development in AI is so fast that even specialists won't make predictions for the future. Firms should not lock-in to specific tools that might not exist in the near future.

Opportunity in Brazil: Brazil's educated workforce and unique challenges present significant opportunities for the use of AI. Brazilians have the hustle and hunger to be amazing entrepreneurs.

TAKEAWAYS BY PANEL:

A Foundation Model Primer: The impact of generative AI - Scott Brady , Innovation Endeavors .

• The pace of tech has changed, adoption is faster than any other in the past (big data, iot, etc.). This speed of change is what makes us uncomfortable.
• The fact that developers were the first to adopt Gen AI for their work created a self-improving cycle for AI, because they were the ones also working to make LLMs better.
• Impact on jobs: Services and personal jobs with human touch won't be affected that much. Changes will happen in mechanical or text intensive jobs such as data entry, analysis, developing.
• Corps are still not ready to put all their data into LLMs. Ideally using sandbox internal models. These problems are talent limited, how many people have the right skills to do it? Brazil has a well-educated workforce that can drive the change.
• Concern with ethics and misinformation: AI can generate hyper targeted fake news to manipulate people with what they are affected the most.

How are corporations embracing AI? A Brazil and US view - Bain AI case team survey 2024 - Bhavi Mehta , Lucas Brossi , 贝恩公司 .

• 87% of USA companies have AI as their strategic priority, while only 48% in Brazil have.
• 58% of companies in Brazil identify as late adopters while USA is only 17%.
• Biggest concern in US is output quality and hallucination and security and privacy, while in Brazil is lack of technical talent.
• Tech firms are using AI for core products and software development. Non tech firms are using for internal productivity, customer service and marketing.
• Adoption of GenAI has the potential to improve EBITDA in 12% on average.
• Back office, sales and marketing can be improved up to 90%. New products and less mechanical jobs are more around 20%.
• 5 to 10% of firms are ready to be bold and not have clear ROI before moving fast and being the first to adopt AI. Best approach is being committed but having a clear vision of where to use AI in the firm.

Reinventing Intelligence: How AI is headed everywhere - Cristiano R. Amon , 高通 . Bianca Martinelli , Alexia Ventures .

• Agrees with Sam Altman that AI is probably under hyped. Developing computing engine on the edge for AI. Devices will be running ai models constantly. Every text, every entry and every interaction will become data and a prompt to the AI models.
• Exciting fields: Smart cameras, energy efficiency, wifi traffic and efficiency, spacial computing.
• Principle of 6G is a sensing network, AI applied to the cellular network for a connected society and intelligent devices, launch in 2030.
• Jobs and work will have to adapt: We can do more with the people we have instead of letting people go. Robots changed factories but didn't end all the jobs. Brazilians are great at improvising, going around obstacles and will adapt fast to this new environment.

Unlocking AI's potential for business and everyone - Fábio Coelho , Philipp Schindler , 谷歌 .

• Brazil has a young, fast moving, risk seeking workforce. Unique challenges that developed countries don't face: Infrastructure, education and health, opportunities to create new companies in those spaces.
• Paradigm shift to AI: Back in the day, Larry and Sergei decided to only look at mobile products when it was still low single digit adoption but this bet proved to be the best when the world became mobile-first. The same thing will happen with AI.

AI Investing Outlook - Michael Nicklas , Valor Capital Group .

Samir Kaul , Khosla Ventures :

• First institutional check at OpenAI and there was no way it could be predicted. AI has been around for long, but it came to the public eyes when consumers started using it.
• Even if for 80% of use cases you can use your own data to train open source LLMs, there's still 20% of a trillion-dollar industry that will use the OpenAI and others.
• You shouldn't make long term predictions today, we have no f* idea what's gonna happen.

Courtney Z. Chow , Battery Ventures :

• Fluency is different from accuracy; one is good for creativity and the other for mission critical uses.
• Supercharging inefficient workflows is the most common use case (co-pilots), but the most interesting ones are those that disrupt the processes themselves. Those are hard to find because the application layer is still too thin in terms of business model.

Mike Ding , Warburg Pincus LLC :

• Warburg is optimistic about Brazil, has invested US$2B in the region so far.
• Previous generations of AI were not ready to deploy, they weren't products, they were internal complex tools.
• Over $100B in combined valuation between the top 3 foundational LLM but they have roughly $1B gross profit. Long term, it’s debatable if they are going to still be the standard. Meanwhile, open-source models seem like good alternatives.
• Will the increase in productivity require people to work even more now that they have the productivity tools at their disposal?

AI Ethical puzzles and frontiers - Rob Reich , 美国斯坦福大学 . Miriam Wimmer , ANPD. Britt H., Salesforce. Gabriel Bayomi Tinoco Kalejaiye , Openlayer .

• Crispr was blocked by the scientific community for human gene designing, would something similar work for AI?
• Analogy: At first, seat belts weren't required on cars. When it became a law many people didn’t agree with it, but now, safety features are actually a differential. After people realize the risks of AI, maybe something similar will happen?
• In Brazil, ANPD already has rules in place for personal data. For AI, the most important thing is individuals having a way to challenge ethical cases, transparency to know what's happening with your data and products you are using.
• Two competing approaches to AI regulation: One very loose that relies on market self regulating, and the other similar to data act, very strict guidelines. Both are competing in congress.
• Something we could try: Risk based approach. Telecom for example is regulated based on the number of clients and size. If you pose a higher risk to people, there are more responsibilities and stricter rules.

AI Showcase Education - Kian Katanforoosh , Workera .

• Half-life of skills: After learning something new, how long does it take before you have to reskill? Before, on the digital space was 4+ years, now every 2 year you need to reskill.
• 10k skills involved in the AI development.
• AI knowledge inside firms is different from cloud for example, because everyone in the company must be an AI person in a different level of skill. However, generic skills or misaligned with the projects are useless. AI learning should be specific to the role and project.

How CEOs should think about AI - Thomas M. Siebel , C3 AI .

• Worked for Oracle when it had 20 employees.
• Our biology shouldn’t be underestimated. Human brain uses very little energy and does computation work equivalent to US$100M worth of GPUs, that are not even able to do 1% of what the brain does in terms of complexity.
• Gen AI models do almost nothing, it's only predicting the next word in a sentence. People don't clearly know how they work, it’s a black box.
• CRMs, ERPs and other corporate software are all good for hindsight. AI makes it possible to predict, forecast, optimize, etc.
• Global AI software revenue in 2027 is expected to be US$1T. GenAI revenue expected to reach US$1.3T in 2032.
• AI is changing the interface between human and tech, it's an abstraction layer. We don't need to train everyone to use AI.
• LLM models are not good enough for widespread corporate use yet: Random responses, no source traceability, no enterprise access controls, cybersecurity risks, prone to hallucination, IP liability, exposure.
• Current large LLM players may die so corporations shouldn't lock in. Yahoo had 95% market share and lost everything. IBM was a leader in cloud and lost the battle.

Game On: The billion dollar untold story - Victor Lazarte , Benchmark . Henrique Dubugras , Brex .

• New platforms start when companies get formed. Development of AI depend on companies being founded around it.
• Most excited about disrupting the computers. Humans had to learn the language of computers and now computers learned the language of humans. GenAI is exactly that.
• Meli and Nubank are the 2 big examples of startups in Latam, and both founders spent time here in the US and brought back the experience. Silicon Valley is an ecosystem based on trust, finding someone who faced a similar problem and is open to helping you is easy.
• There are companies that can be created anywhere and others that must be born in specific places. You can't start the new LLM company based in Brazil because you need the best talent. But you can tackle specific challenges of the region by being based there. Application layer solutions can be huge in Brazil for example.
• Benchmark: Business of finding exceptions, waiting to be surprised. Best Founders = Open minded, organized, disagreeable (stubborn). Hard to find stubborn, you must not be relaxed when you are sitting with that person, it's counter intuitive.

AI Powered Fintech: A founder's journey - Pedro Franceschi , Brex . Anu Hariharan , Avra.

• Started jailbreaking iPhones in Brazil at the age of 12, saying he was 14 to be able to work with a startup. TED Talk at the age of 13. After Pagar.me , started Brex with the idea of a prepaid debit card that worked as a credit card.
• Frequent questions to leaders at Brex: What is the company bottleneck? What is one thing that is limiting YOUR growth? Usually break all the rules in the company to get things done and breaking the barriers to let people run.
• Most fundamental questions for startups, regardless of AI: Do you understand the customer? Can you build something to help them? and will they pay for it?
• Hiring in BR: Great talent is everywhere, no difference in IQ between US and BR, but difference in training, so had to calibrate the interviewing process. Had very bright people who failed the processes and then helped candidates prepare for US interviews. Once inside, hustle and hunger were much larger than US employees, and culturally this incentivized the entire team.