Foundations for a Better World - MIT AI Conference 2023 ✨ (Part II)
Foundations for a Better World - MIT AI Conference 2023 ✨ (Part II)
The MIT AI Conference happened on 10/21, with speakers like Sal Khan, Greg Brockman, Sam Altman, Dava Newman, and more
Duanduan had the great pleasure volunteering at the MIT AI Conference at the Computer History Museum in Mountain View on 10/21. The theme for this year is “Foundations for a Better World - Shaping The Future With Responsibility, Creativity & Collaboration.” The panels are structured to reflect the areas that are at the forefront of AI revolutions: healthcare, transportation, asset management, enterprise, trust & safety, security, supply chain…
Below are some of the insights fresh from the conference (Part II). 🤩
Link to Part I:
🔗 Here is the list of speakers at the conference (in no particular order).
We have also summarized a lot of exciting deals this week! So be sure to check them out as well.
💬 Fireside Chat with Sam Altman & Daniela Rus: AI at the Cutting Edge
Sam Altman: OpenAI / Daniela Rus: MIT-CSAIL / Moderated by: Teddy Lee (OpenAI)
Societal Impacts of Advanced AI:
Daniela is optimistic about AI's potential benefits, such as improved disease diagnosis, personalized medicine, reduced road fatalities, tailored educational programs, enhanced privacy, efficient transportation, understanding of climate and planet, and increased accessibility for people with disabilities.
Sam sees AI as a tool for amplifying human capabilities. He believes AI will significantly boost productivity and foster advancements in science.
Risks of AI Development:
Sam is concerned about humans misusing AI, including for weaponization. As AI systems become more powerful, he also worries about the sci-fi-like risks they might pose.
Daniela emphasizes the dangers of misinformation and disinformation, especially as AI becomes better at producing realistic fakes. She stresses the need for technical solutions, like watermarking technologies and ensuring the trustworthiness of information sources.
Open vs Closed Source:
Daniela advocates for open source as a means to understand and improve transparency in AI models. She mentions that open sourcing can provide insights into how large models operate, and why hallucinate. But open-source solutions also empower supervillains, and we need to keep that in mind.
Sam acknowledges the importance of open source but highlights the need for balance between proprietary and open technologies.
Evolution of Autonomous Agents:
Sam believes people desire AI agents that can communicate and act autonomously. These agents will start with limited capabilities and gradually gain more as risks are understood and mitigated.
Daniela envisions agents assisting with both cognitive and physical tasks. She sees the future involving improved interactions between AI and the physical world, leading to agents with better human understanding.
Implications for the Job Market:
Daniela is optimistic about AI's potential to assist and augment human roles. She mentions a study where AI and doctors combined reduced diagnostic errors. But low-level routine jobs will be automated, so it’s important to prepare by teaching the new tools and elevating the work of people.
Sam agrees that while AI will create new jobs and improve existing ones, it will also lead to job losses. It is going to be a disruption to people’s lives. He emphasizes the importance of empathy in the industry towards those affected by job displacements. On the positive side, the amount of work each of us can do, the leverage on creative spirit and ideas, and the new opportunities that are being created are amazing.
A recent MIT study found that the current cost of implementing certain automation tasks, like visual inspection, is prohibitive. Thus, society has some leeway before grappling with AI-driven inequalities.
To watch the full video for this panel, please check out Teddy’s YouTube:
🔐 AI in Cybersecurity: Defenders’ Dream or Nightmare?
Rohit Ghai: RSA / Steve Wilson: Exabeam / Yousuf Khan: Ridge Ventures / Moderated by: Rami Elkhatib (Acero Capital)
3 Dimensions of AI in Cybersecurity:
AI as a Shield: AI promises significant advancements in protective technologies.
AI as a Weapon: Not all hackers are skilled programmers; many use tools created by a select few. With AI, these cyber weapon creators have the potential to develop even more potent threats.
AI as a Target: New technologies invariably bring new vulnerabilities. Post the launch of ChatGPT, threats such as prompt injection and data poisoning have become areas of concern.
Security in Enterprises:
Yousuf reflected on his 5 times as CIO, the number 1 threat to enterprise security is phishing, with about 70% of phishing traffic being automated and frequently mutating. It’s becoming more sophisticated and targeted.
Security is everyone's concern, from the board level to employees. The motive behind many cyberattacks is not always data theft but to create chaos, as cyber warfare gains prominence.
A balance must be struck between fostering innovation and maintaining a robust cybersecurity posture.
Zero trust is a prevalent approach in cybersecurity that works on the principle of 'least privilege'. This means that every actor in a network should have minimal privileges, only as much as they need at that particular moment. Given the complexity of modern IT landscapes, it's challenging for humans to manage this intricate mapping of privileges. AI can facilitate the implementation of the zero-trust approach by crafting nuanced policies that define who has access to what resources.
Talent Gap in Cybersecurity & Role of AI in Filling the Gap:
There's a massive shortage of professionals in cybersecurity, with millions of positions unfilled. The unfilled positions have increased by 26% in just a year. AI is positioned as a solution to this talent gap.
There are different tiers of cybersecurity professionals, with tier-three analysts being the most skilled. There's a shortage at the tier-one level. This gap can be filled with AI since many tasks at this level are automatable.
Future of AI and Cybersecurity:
AI's role in cybersecurity is not limited to just defense but also spans to securing AI infrastructures. For instance, ensuring the privacy and compliance of data, protecting machine learning models, and guarding the code at the chip level.
Nation-state actors play a significant role in the global cyber threat scenario.
The panel advocates for SaaS companies to utilize existing frameworks and guidelines like those from CISA and NIST to improve their cybersecurity posture.
🧑🚀 Next Generation Foundation Model, Human-centric AI
Surya Ganguli: Stanford / Moderated by: Bridget Brett (LiveRamp)
AI and Science:
AI possesses the remarkable capacity to uncover intricate patterns in data, placing it on par with groundbreaking technologies such as the microscope. Surya is especially excited about AI shedding light on human biology and analyzing brain patterns.
Limitations of Current AI Systems:
Surya discussed the marvels of AI advancements like GPT-4 but also highlighted their limitations. While AI can ingest enormous amounts of data, humans are more energy-efficient and less prone to specific types of errors that can fool AI. Surya perceives AI as a "different type of alien intelligence."
AI Ethics and Society:
Regarding societal rules for AI, Surya referred to Asimov's laws as a starting point but insufficient.
Open democratic discussion about AI's societal impacts should be advocated. A few big tech companies shouldn't singularly determine AI's trajectory in our lives. Surya championed a democratic approach to regulating and controlling AI to ensure a harmonious future between humans and AI systems.
AI “Doom”: Surya dismissed the extreme fears of AI leading to human extinction as overblown. While AI-human interactions need careful thought, concerns about more immediate societal issues, like climate change or democracy's future, are more pressing.
🏛️ Applications of Foundation Models
Arvind Jain: Glean / Sami Shalabi: Maven AGI / Cai GoGwilt: IronClad / Moderated by: Calvin Chin (E14 Fund)
Foundation models as Enablers, not End Goals:
Both incumbents and startups should view foundation models as tools that enable a larger vision or solve a critical pain point. They shouldn't be seen as the only value proposition.
Sami highlighted that Maven AGI didn't begin with the intent of just using GPT-4; the company wanted to solve a problem and found that the model was the best tool for the task.
Startups vs. Incumbents:
Incumbents might have an Edge in Adoption: Arvind noted that large companies like Google, Microsoft, and Salesforce can rapidly enhance their existing products by seamlessly integrating these models. This can lead to immediate improvements in user experience and, given their existing large user base, can mean rapid adoption and feedback.
Cultural adoption of AI in enterprises is a challenge: AI induces fears around job displacement or concerns about data privacy and security. Engaging and educating not just the technical teams, but also the broader employee base, becomes crucial. Demonstrating tangible benefits, ensuring transparent communication, and providing continuous training can help ease such transitions.
Startups Need to Differentiate: Relying solely on foundation models for differentiation might be a weak strategy, especially given that many companies can access the same technology. For startups, it's essential to combine the power of these models with unique solutions to existing problems, a distinctive user experience, or a novel application.
Take a Swing at a Home Run: For companies that have been around for a while, it might be worthwhile to take more significant risks by attempting to solve bigger challenges using these models. Cai's advice to "go for a home run" speaks to this, highlighting that for established companies, incremental improvements might not be as valuable as transformative changes.
Production Quality Systems: Building a production-quality system requires more than just coding the main functionality. It includes monitoring, observability, security, and more, which often demands additional education and understanding.
Future of Support and Contracting:
Both the cases of Maven AGI and IronClad highlight the transformative potential of AI in areas like customer support and contract negotiation. The ability to automate, personalize, and enhance such traditionally manual and tedious processes indicates a significant shift in how businesses will operate in the future.
Job Security and AI:
There's a consensus that companies refusing to adopt AI will lag behind, and employees should advocate for AI adoption. Those who don't embrace AI risk being left behind in the workforce.
🔥 AI Deals of the Week
Databricks Closed Series I with Additional Strategic Investors
Keywords: Data Lake, AT&T, AWS, Microsoft
Databricks closed its Series I with additional investments from AT&T Ventures, QIA, Sanabil Investments, AWS, CapitalG and Microsoft. According to a previous announcement, the company is raising $500M at $73.50 per share and a post-money valuation of $43B.
Green Street Acquired Local Data Company
Keywords: Commercial Real Estate
Green Street, a commercial real estate intelligence and analytics company, acquired Local Data Company, which provides data on retail and leisure real estate markets the Great Britain.
Aleph Alpha raised $500M in Series B
Keywords: Model, B2B
Aleph Alpha raised $500M in the round led by Schwarz Group and Bosch Ventures.
May Mobility Raised $105M in Series D Funding
Keywords: Japan, Autonomous Vehicle
NTT Group led the Series D of May Mobility, an autonomous driving company focusing on B2B and business-to-government. NTT Group has acquired the exclusive rights to distribute May Mobility’s proprietary autonomous vehicle technology throughout Japan, and May Mobility will work with Toyota to develop autonomous driving ecosystem.
Kythera Labs Announced $20M Series A and Debt Funding
Keywords: Healthcare
Kythera builds a healthcare data platform, Wayfinder, with access to real-time healthcare data over 8 years. The platform is built on Databricks. The funding is provided by BIP Ventures and CIBC.
Certn Acquired Trustmatic
Keywords: Fraud Detection, Security
Certn, a background screening company, acquired Trustmatic, a company based in Bratislava providing remote identity verification and fraud detection. Founded in 2020, Trustmatic has previously raised a pre-seed round led by Neulogy Ventures.
IBM Unveiled $500M Enterprise AI Venture Fund
Keywords: Generative AI, Fund, IBM
IBM established a new $500M fund focusing on using Generative AI for enterprises and strategic partnerships between start-ups and IBM, utilizing IBM’s watsonx platform.
BioPhy Raised $4.5M in Pre-Seed
Keywords: Healthcare, Clinical Trials, Drug Discovery
BioPhy is a clinical trial data company. Offered by BioPhy, BioLogicAI predicts the successfulness of a drug or the outcome of clinical trials. Another product, BioPhyRx, uses generative AI to help successful drugs go to market. The round is backed by Metrodora Ventures and individual investors.
Playbook Acquired CanFY
Keywords: Media Cloud Storage, AI Assistant, Image Recognition
Playbook, a visual media storage and collaboration company, acquired CanFY, a company building AI assistant. The acquisition is intended to boost the visual search capability of Playbook. Playbook raised Series A led by Bain Capital Ventures in 2022.
Luniko Technologies Raised $700K in Non-Dilutive Pre-Seed Funding
Keywords: AI Adoption, Digital Transformation
Luniko Technologies helps Canadian companies adopt AI into their operations.
Black Ore Emerged from Stealth and Raised $60M in Seed
Keywords: Fintech, Tax Return, US
Black Ore is a fintech company that develops Tax Autopilot, a SOC-2 compliant tax filling platform to speed up some of CPA’s work from days to seconds. The round is led by Oak HC/FT and a16z, with participation from other prominent funds and individuals.
Protecto Raised $4M in Seed
Keywords: Cybersecurity, Data Security
Protecto provides AI privacy protection solutions for financial services, healthcare and tech industry enterprises. It provides a suite of products ranging from PII redaction to data privacy compliance. The seed round is led by Together Fund.
Osmo Received a $3.5M Grant
Keywords: Insect Control, Chemistry
Osmo discovers insect repelling molecules to control disease carrying insects. It received the grant from Bill & Melinda Gates Foundation to advance its AI-enabled scent discovery process.
Tabnine Raised $25M Series B
Keywords: Code, Automation, Dev Tools
Tabnine provides a subscription service and VSCode plugin for code auto-generation. Telstra Ventures led the round, with participation from Atlassian.
Ghost Autonomy Raised $5M
Keywords: Autonomous Vehicle, Multi-Modal AI
Ghost Autonomy brings multi-modal large language to autonomous driving softwares for consumer vehicles. It may lower the hardware requirements to democratize self-driving cars. The funding is provided by OpenAI Startup Fund.
Eleos Health Raises $40M in Series B
Keywords: Behavioral Health, Healthcare
Eleos collects in-session client conversation data and turns it into insights for better patient experience. The round is led by Menlo Ventures.
Eilla AI Raised $1.5M in Seed
Keywords: Due Diligence, Document Research
Eilla AI uses generative AI to assist M&A, VC and PE with researches. Eleven Ventures led the round.
Flip AI Announced a Previous $6.5M Seed Round
Keywords: Observability, DevOps, IT Infrastructure
Flip AI brings AI into DevOps and helps developers collect observability data through AI-reasoned debugging steps and auto-generated queries. Factory led the round with participations from Morgan Stanley Next Level Fund and GTM Capital.
Pioneer Raised $2.9M in Seed Funding
Keywords: Government Funding, Fundraising
Pioneer helps Climate Tech start-ups find non-dilutive fundings from governments by utilizing LLM. Blue Bear Capital has led this seed round.
Evolv AI Announced $13.3M in Funding
Keywords: UX, User Analytics
Evolv AI provides user design and user behavior analytics to facilitate user experience improvements. The funding is a combination of the conversion of previously issued convertible debts and new equity. It has raised $23.3M funding to date.
Terray Therapeutics Announced Funding from Nvidia and Others
Keywords: Drug Discovery, Healthcare
Terray Therapeutics uses generative AI to improve the speed, cost, and success rate of small molecule drug discovery and development. It would leverage Nvidia’s DGX Cloud to develop chemistry foundation models for small molecules. It received investment from NVenture, NVIDIA's venture capital arm, and other investors.
DeepInfra Emerged from Stealth and Received $8M in Seed Funding
Keywords: CPU, Low Cost Inference
DeepInfra provides ways to deploy foundational models on CPU and other lower-cost hardwares. However, its website shows that it currently provides LLM API endpoints running on Nvidia A100 GPUs. The funding round is led by A.Capital and Felicis.
Aigen Raised $12M in Series A Funding
Keywords: Agriculture, Clean Energy, Robotics
Aigen raised $12M Series A led by ReGen Ventures. It builds solar-powered autonomous robots for crop management without chemicals. The fund will be used to build a 7,500 sq. ft. manufacturing and R&D facility to manufacture their solar-powered robotic fleet.