• Project Flux
  • Posts
  • Surpassing Humans: The Quest for Safe Super Intelligence

Surpassing Humans: The Quest for Safe Super Intelligence

We haven’t even reached AGI but the mission to reach safe super intelligence has started. Will AI really surpass us sooner than we think?

Author

James Garner
June 24, 2024

Opinion

As the race towards artificial general intelligence continues, questions are asked regarding the ethics and safety of this cognitive-industrial revolution.  Former Open-AI co-founder Ilya Sutskever has announced a rival startup looking to surpass the blazing trail left by OpenAI.  Ilya has set a mission to reach safe super intelligence (SSI), which is also the name of his new startup showing his dedicated commitment to the cause. But what is safe super intelligence?

Super intelligence refers to an AI system that surpasses human intelligence across a wide range of cognitive abilities. Ilya’s believes achieving this safely is the “most important technical problem of our time”.

We find this announcement to be a breath of fresh air. Firstly because SSI are approaching safety and advanced capability in tandem, showing equal priority. Secondly, with so many businesses leveraging AI for commercial purposes, SSI have “one goal and one product: a safe super intelligence”. Their singular focus unburdens them from commercial pressures meaning they can focus on building intelligence which can scale safely. We do have some skepticism however, especially seeing how OpenAI started as a non-profit to which later became a for-profit with Ilya amongst the decision makers. Time will tell but the signs are promising.

Graph of the Week

This graph illustrates the escalating costs of training AI models from 2017 to 2023. Each bubble represents an AI model, with the size reflecting the training cost. The Gemini Ultra, trained in 2023, is the most expensive at $191.4 million. Other significant costs include GPT-4 at $78.4 million and PaLM 540B at $12.4 million.

Earlier models like BERT-Large (2019) and Transformer (2017) had much lower costs, $3.3k and $930, respectively. The graph underscores the rapid increase in AI training expenses over time, influenced by factors such as hardware value, utilization rate, and training duration. Major contributors to these models are Google, OpenAI, Meta, Microsoft, and NVIDIA. Data is sourced from the Stanford University Artificial Intelligence Index Report 2024. This trend highlights the growing complexity and resource demands of modern AI development.

Quote of the Week

There is a big difference between tech as augmentation versus automation. Augmentation (think Excel and accountants) benefits workers while automation (think traffic lights versus traffic wardens) benefits capital.

LLMs are controversial because the tech is best at augmentation but is being sold by lots of vendors as automation.

This Week's Big Thing

Nvidia Becomes the World's Most Valuable Public Company

In a historic shift, Nvidia has surpassed Microsoft to become the world's most valuable public company. As of mid-June 2024, Nvidia's market capitalisation soared to $3.33 trillion, fuelled by its dominant position in the AI chip market. This leap was driven by a 170% rise in Nvidia's stock since the beginning of the year, reflecting strong demand for its AI chips as major tech firms race to advance their AI capabilities​ (Benzinga)​​ (Summa Money)​.

Nvidia's success is closely tied to the burgeoning AI industry, with its chips being integral in data centres and AI model training. The company now holds about 80% of the market share for AI chips in data centres, a sector that saw a 427% revenue increase year-on-year, reaching $22.6 billion last quarter​ (Summa Money)​. This growth not only propelled Nvidia's stock but also positioned it ahead of Apple, which it surpassed earlier in June​ (Ground News)​.

The rapid ascent of Nvidia's market value underscores the transformative impact of AI technology on the tech industry. Nvidia's strategic investments and innovations in AI have paid off, making it a cornerstone in the AI ecosystem. With significant market share and continuous innovation, Nvidia's rise exemplifies the potential and profitability of AI-driven technology in today's market​ (Ground News)​.

For more details on this landmark achievement and its implications, you can read the full articles on Benzinga, Ground News, Summa Money, and CNBC.

Tip of the Week

Personalise Your ChatGPT Instructions

Enhance your ChatGPT experience by personalising its instructions to better suit your needs. Here are some tips on how to customise your settings:

1. Access Settings: Click on your profile, go to settings, and select "Personalisation."

2. Define Your Tone: Adjust the tone and style of responses to match your preferences. For instance, you can make the language more formal or conversational.

3. Set Specific Guidelines: Add detailed instructions on how you want responses to be framed. Include preferences on brevity, clarity, and whether to break down complex tasks.

4. Feedback Loop: Regularly update your instructions based on how well they are working for you.

Here are some example instructions you can use:

· "Always respond in British English."

· "Keep answers concise and to the point."

· "Break down complex problems into smaller, manageable steps."

· "Avoid expressions of remorse or apology."

· "Provide multiple perspectives or solutions where applicable."

By tailoring these settings, you can ensure that your interactions with ChatGPT are aligned with your specific needs and preferences.

AI in Project Delivery

  • Network Rail Revolutionising Knowledge Management through Generative AI
    Network Rail leverages generative AI to enhance knowledge management, improving efficiency and decision-making in project delivery. Read more here.

  • BMO Leverages Matterport’s Digital Twins to Streamline Acquisition Work
    BMO uses Matterport's digital twin technology to streamline real estate acquisition processes, enhancing accuracy and efficiency. Discover the details here.

  • Geoffrey Hinton's New Venture: AI and Carbon Capture
    AI pioneer Geoffrey Hinton, after leaving Google, joins CuspAI to work on AI-driven carbon capture solutions. Learn more here.

  • Citi Predicts AI Will Displace More Finance Jobs Than Any Other Sector
    Citi's report highlights AI's potential to displace finance jobs significantly, reshaping the industry landscape. Read the full story here.

  • United in the Delivery of Responsible AI
    Digital Construction Today discusses the importance of responsible AI in construction, advocating for ethical standards and practices. Find out more here.

  • AI in Construction Embraced in the UK Despite Challenges
    The UK construction industry adopts AI technologies, overcoming challenges to enhance project delivery and efficiency. Read more here.

  • Automation Demands Drive Boom in Global Smart Buildings Market
    Rising automation demands are fueling growth in the smart buildings market, revolutionising construction and facility management. Discover more here.

Tech News

  • DeepMind's VEO Technology
    DeepMind introduces VEO. Learn more here.

  • Microsoft Delays AI Recall Feature for Windows
    Microsoft postpones the release of its AI Recall feature for Windows, aimed at improving task management and productivity. Details here.

  • Anthropic's Claude 3.5 and Sonnet Launch
    Anthropic unveils Claude 3.5 and Sonnet, its latest AI models designed to advance language understanding and generation. Read more here.

  • Apple Enhances AI with 20 New Open-Source Models
    Apple researchers add 20 new open-source models to improve AI capabilities in text and image recognition. Discover more here.

  • The Cost of AI Compute: Epoch Report
    An in-depth analysis of the escalating costs associated with AI compute, detailing financial and environmental impacts. More information here.

  • Roblox's Journey to 4D Generative AI
    Roblox outlines its roadmap to integrating 4D generative AI, aiming to revolutionise user experiences on its platform. Learn more here.

  • Apple to Pay OpenAI for ChatGPT via Distribution
    Apple will compensate OpenAI for ChatGPT through distribution methods rather than direct payments. Full story here.

  • Apple Pauses High-End Headset Development
    Apple suspends work on its next high-end headset to focus on a cheaper model expected to launch in late 2025. Details here.

  • Google's Green Light Traffic Optimisation
    Google introduces AI-driven traffic light optimisation to reduce congestion and improve traffic flow. Learn more here.

  • RunwayML's Generative Video Tools
    RunwayML launches new generative video tools to enhance creativity and production efficiency. Discover more here.

  • SoftBank CEO Predicts Advanced AI
    SoftBank CEO predicts future AI will be 10,000 times smarter than humans, highlighting AI's exponential growth. Full story here.

AI Security News

  • McDonalds : AI Surveillance Concerns
    The BBC discusses the recent announcement that McDonalds are pausing their AI service….for now. Read the full article here.

  • General Intelligence 2024: AI Safety Summit
    A recent summit on AI safety in 2024 highlighted advancements and concerns in AI development, emphasising the need for robust safety measures. Learn more here.

  • Tech Company Enhances AI Security
    A tech company has laid off 60 workers and replaced them with Chat GPT. Find out more on LinkedIn here.

  • Apple Delays AI Rollout in Europe
    Apple delays the rollout of its AI initiatives in Europe due to regulatory and safety concerns. Details available here.

New Tools in Tech

FinalSpark's NeuroPlatform

FinalSpark's NeuroPlatform enhances productivity with AI-driven data analytics and automation. It integrates data sources, offers predictive analytics, automates workflows, and features customizable dashboards and collaboration tools. Ideal for healthcare, finance, and manufacturing, it increases efficiency and improves decision-making. Explore more here.

Event of the Week

nPlan AI Day - Summer 2024

Date: Thursday, 27 June 2024
Time: 4:30 pm (doors open at 3:45 pm)
Location: Fora - Broadwick Street, 33 Broadwick St, London W1F 0DQ
Format: In-person and online

About nPlan: nPlan is a cutting-edge company that uses AI to predict and manage risks in project planning and delivery. Their technology leverages vast amounts of data to optimise construction projects and minimise delays.

Event Highlights: The AI Day will showcase nPlan's latest innovations, including advancements in their flagship tool, "Barry," which aids in project planning by predicting potential issues before they arise. The event will feature live demonstrations, expert talks, and networking opportunities.

How to Attend: Register to attend either in-person or virtually by visiting nPlan AI Day.

Podcasts: Must-Subscribe List

In Lex Fridman Podcast #434, Aravind Srinivas, CEO of Perplexity, delves into how his company is transforming the way we find answers on the internet. Perplexity leverages a combination of search engines and large language models (LLMs) to produce highly accurate answers, with every part of the response citing human-created sources on the web. This approach significantly mitigates the issue of LLM hallucinations and enhances reliability for both research and casual inquiry.

Aravind, who has an impressive background as a former PhD student at Berkeley and an AI researcher at DeepMind, Google, and OpenAI, discusses various technical aspects of machine learning and innovations in retrieval-augmented generation (RAG). He also explores chain-of-thought reasoning, the process of indexing the web, and the importance of user experience (UX) design in developing effective AI systems.

The conversation covers the state-of-the-art in machine learning and touches on how Perplexity's methods improve the accuracy and trustworthiness of information retrieval. Aravind emphasizes the potential for Perplexity to revolutionize internet searches, making it a valuable tool for both detailed research and late-night curiosity-driven explorations.

The podcast also features brief mentions of sponsors like Cloaked for cyber privacy, ShipStation for shipping solutions, NetSuite for business management, and LMNT for electrolyte supplements. These sponsors help support the podcast and provide valuable services to listeners.

For anyone interested in AI, search technologies, or the future of internet information retrieval, this episode offers a wealth of insights and technical details. Listen to the full episode on Apple Podcasts here.