This Week’s BIG thing

Tesla’s We-Robot: Autonomous self-driving cars and robot waiters

Last week, Tesla hosted its highly anticipated “We Rebot” event, showcasing, you guessed it... advancements in robotics and autonomous vehicles. To many, the unveiling was revolutionary, but to some it lacked depth creating a divide amongst investors and analysts. Let’s dive in.

 Key Announcements

1.     Cybercab and Robovan unveiling
Tesla introduced the Cybercab, a two-seater autonomous vehicle without a steering wheel or pedals. It is set to be under $30,000 with production anticipated to start "before 2027." They’ve also unveiled the Robovan, an autonomous shuttle designed to transport up to 20 passengers or cargo. Though no pricing or production dates were shared, it hints at Tesla's ambition to revolutionise public transport.

2.     Full Self-Driving (FSD) Updates
Unsupervised FSD features will be rolled out in Tesla Model 3 and Model Y vehicles across California and Texas by next year, predicting these autonomous vehicles would be “10 to 20 to 30 times safer than human drivers.”

3.     Optimus Robot Demo
Roughly 30 humanoid robots made a lively appearance, mingling with guests, serving drinks, and even performing a dance. Whilst impressive, the AI capability on show was questionable...

AI Capabilities on Display

Tesla’s AI played a key role throughout the event. The company's advancements in autonomous driving continue to hinge on a camera-based vision system that mimics human eyesight, avoiding the use of more traditional sensors like LiDAR. This approach is faster, simpler, cheaper, more scalable (as it doesn’t require pre-mapping) and most of all, bio mimics the marvel that is human vision. This approach however, is a single point of failure because of insufficient redundancy in comparison to LiDAR, it is limited by certain weather and lighting conditions, and has a limited range. It’s worth noting Waymo use LiDAR tech and have achieved autonomous driving in certain areas before Tesla.

Main criticisms were that despite having autonomous humanoid robots mingling with people, some bots admitted to being remotely assisted by humans. This has raised questions into the authenticity of AI capability on show, particularly because of the lack of detail shared by Tesla on AI developments. This was a theme across the whole show whereby the event was heavier on the wow factor and lighter on the business and technical details cared for by investors and analysts. Despite bold promises, investors left concerned due to the absence of clear timelines, financial projections, or details about the robotaxi fleet’s rollout. This contributed to an 8.8% drop in Tesla’s stock, wiping out $67 billion in value. Other criticisms include regulatory hurdles from no peddles or steering wheels, safety concerns and over-ambitious timelines too often promised by Musk.

Why this matters 🫵

Tesla’s ambitions are ambitious and it’s no surprise. Imagine a world where your vehicle can drive you anywhere without your intervention, or even better, it can run errands for you whilst you lounge in bed…or better yet it can self-park and be cleaned by your humanoid AI robot!? Musk is creating an autonomous travel ecosystem. Despite schedule setbacks and critiques, Musk is known for setting course in uncharted territory and has delivered in other areas such as the recent SpaceX breakthrough, so we’re cutting him some slack. Either way, we have a feeling this unique event will be spoken about for years to come.

The rabbit hole 🐰

_{Watch the full event} 🔗

_{A deeper look on worried investors} 🔗

_{Optimus robots admitting human assistance} 🔗

_{Watch the Tesla reveal in 8-minutes} 🔗

What’s new: Tech

AI isn’t as smart as we thought?!

There’s been a shocking breakthrough questioning the reasoning power of large language models (LLMs) that power tools like ChatGPT. Despite their apparent sophistication, recent studies suggest these AI systems might be relying more on clever pattern matching than genuine logical thinking and reasoning. This insight shakes up the assumption that LLMs have human-like cognitive abilities and raises big questions about their dependability, especially in solving complex problems. AI might not take our jobs after all?

Apple's New Benchmark: GSM-Symbolic

Apple's research team has stepped up the scrutiny with a new benchmark called GSM-Symbolic. This test goes beyond the usual, aiming to put LLMs’ mathematical reasoning under a more intense spotlight. The results? Not exactly what we were hoping for. Here are the key findings:

1.     Inconsistent Results: LLMs often gave varying answers to the same question posed in slightly different ways, showing a lack of consistent reasoning. For instance, just changing a person's name in a question lowered scores...

2.     Numerical Sensitivity: When researchers changed only the numbers in a question, the models’ performance dropped, pointing to a reliance on memorisation instead of true understanding.

3.     Trouble with Complexity: As questions became more complex, with additional steps or clauses, LLMs struggled, unable to navigate multi-step reasoning effectively.

4.     Easily Distracted: When irrelevant but seemingly related information was added to a problem, performance plummeted by up to 65%, revealing a major flaw in distinguishing between what's important and what isn’t. Here’s an example:

• a.     Oliver picks 44 kiwis on Friday. Then he picks 58 kiwis on Saturday. On Sunday, he picks double the number of kiwis he did on Friday, but five of them were a bit smaller than average. How many kiwis does Oliver have?

• b.     Most humans know that the size of the kiwi’s is irrelevant, but the AI often considered this in its calculations leading to mistakes, not so smart ey?

What’s killing AI reasoning?

Content. Both humans and LLMs show similar behavior when it comes to reasoning influenced by content. But this comparison doesn’t exactly flatter the AI models. Here’s why:

1.     Semantic shortcuts Both humans and LLMs tend to perform better when the problem’s content guides them toward the right answer, suggesting a heavy reliance on prior knowledge, not logical deduction. AI is more sensitive to this

2.     Struggle with abstraction Like humans but more so, LLMs stumble when faced with abstract or unfamiliar scenarios, showing a reliance on transfer learning, and that content-independent reasoning remains a challenge.

3.     Lack of learning pathways: While humans can improve their abstract reasoning through education, LLMs don’t yet have a clear way to evolve their understanding with meta-learning still in its infancy.

What this means for AI’s future

The limitations exposed by this research carry significant implications for how we develop and use AI technology:

1.     Trust issues: The inconsistency of LLMs in reasoning tasks casts doubt on their reliability in critical areas like education, healthcare, and decision-making. It’s worth noting that smaller models struggled the most whilst OpenAI’s o1 model had an ~18% performance drop off.

2.     Overhyped capabilities: Current benchmarks may be inflating expectations, causing us to overestimate what these models can really do.

3.     Data contamination is at play: The findings suggest that it’s likely that the models’ training data was contaminated with answers from benchmark question sets such as GSM8K, implying that they are just memorising answers rather than logically solving them.

4.     A need for fresh thinking: Despite claims from giants like Sam Altman, simply increasing the size of data or computing power won’t magically fix these reasoning issues. We need new methods to develop AI that can genuinely think, not just regurgitate.

5.     Ethical dilemmas: Relying on pattern recognition over true reasoning raises important ethical concerns, particularly in high-stakes environments requiring critical thought.

Why this matters 🫵

While LLMs are undeniably impressive in many areas, their ability to reason like humans is still very much a work in progress. As we continue to embed AI deeper into our lives, it’s crucial to confront these limitations head-on and better understand how LLM’s operate. After all, planes don’t fly like birds, boats don’t swim like fish and LLM’s don’t reason like a human. Learn how to use AI to augment what you do, not automate the high-value activity that you do well.

The rabbit hole

The viral tweet that kicked it off 🔗Read the breakthrough paper 🔗

A great video walkthrough of the paper 🔗

💙 Other tech news we’re loving

• The 2024 Nobel Prize in Physics honours breakthroughs that bridge quantum science and artificial intelligence, awarded to Pierre Agostini, Ferenc Krausz, and Anne L’Huillier for advancements in attosecond physics. Their work on ultrafast electron dynamics opens unprecedented insight into atomic-scale processes, promising transformative applications in quantum computing and materials science. This year’s awards also reflect the deepening impact of AI on scientific discovery. Geoffrey Hinton and John Hopfield, celebrated for their foundational work in neural networks, were recognised alongside chemists David Baker, Demis Hassabis, and John Jumper for their achievements in protein folding using AI, which could accelerate drug discovery and revolutionise biotechnology. Hinton and Hopfield’s early contributions paved the way for today’s powerful machine learning applications, establishing AI as a cornerstone in modern research.  This Nobel recognition underscores how AI is reshaping fundamental research, enabling discoveries in physics and biology that could redefine scientific frontiers while prompting discussions on the ethical implications of machine learning in science. For more details, see the official announcement here and insights from related coverage across BBC, CNN, and The Verge.

• Meta’s Yann LeCun dismisses fears of superintelligent AI as "complete B.S.," suggesting we are far from creating AI that could pose an existential threat. Read more.

• Google’s new AI image generator, Gemini, is now widely available for free but comes with a limitation: users can only generate a limited number of images per day. Learn more.

• AMD launches a new AI chip, aiming to compete with NVIDIA in the data centre market, underscoring the increasing hardware race in AI technologies. Explore more.

• Meta AI expands into new countries, launching in Brazil, the UK, and more, as Meta pushes its AI presence across global markets. Read more.

• OpenAI’s "Search GPT" is expected later this year, but the company won’t share ad revenues with websites, sparking discussions around fair value sharing. Learn more.

• Google is adding ads to AI-generated overviews, showcasing a new way to integrate monetisation into AI-organised content pages. Explore more.

• Meta's new "MovieGen" text-to-video tool could rival Suno, creating AI-generated videos from text prompts, expanding media capabilities in generative AI. Read more.

• Apple’s "Apple Intelligence" will roll out on 28th October, promising new AI-driven features to enhance user experience on Apple devices. Discover more.

What’s new: Projects

Buildots and Autodesk BIM 360: Progress in Field Management or a Step Toward a ‘Walled Garden’ in Project Delivery?

Buildots has announced an integration with Autodesk BIM 360, streamlining field management and automating issue creation for construction teams. This partnership, detailed here, enables users to track site progress using Buildots’ AI-driven technology within Autodesk’s widely adopted ecosystem, enhancing real-time data flow and project oversight.

While this integration promises efficiencies, there’s a growing concern that big tech players like Autodesk could dominate the market, creating a “walled garden” effect in project delivery software. As more solutions intertwine with industry giants, construction and project delivery risk facing a “Spotify or Netflix moment” – where choices are constrained by a handful of dominant platforms. While the Buildots-Autodesk partnership exemplifies progress, industry professionals must stay vigilant. Open ecosystems and balanced competition are essential to prevent a monopolised digital landscape that could prioritise tech giants over diverse, adaptive project delivery solutions. 

Integrating AI in business requires a people-centric approach, focusing on realistic expectations, leadership support, and staff readiness. Leaders should understand AI's capabilities, foster a learning culture, and offer training. A strategic guide can aid with examples, practical steps, and case studies to drive thoughtful, effective AI transformation.

💙 Other Project news we’re loving

• Nemetschek’s new AI layer across its solutions enhances construction workflows with advanced automation and predictive insights, underscoring the importance of a people-centred approach for effective AI integration in project delivery. Read more.

• AI-driven platforms like ChatGPT are raising concerns about election interference, as their influence on public opinion introduces new risks of manipulation and misinformation during global elections. Explore more.

• Amazon’s AI-enabled Ring doorbells could hint at the future of construction site security, offering new video search features that have shown mixed results in terms of effectiveness. Learn more.

• Wimbledon will replace line judges with AI starting next summer, boosting accuracy and reducing player disputes, which marks the end of a 147-year tradition of human line judging. 
  

What’s new: Productivity

Exporting Expertise: How Personal AI Models Could Transform Project Delivery

In a forward-looking piece from Andreessen Horowitz, “Export Yourself to AI” explores the transformative concept of capturing individual knowledge and skills in digital form. By "exporting" oneself to AI, professionals can create personalised AI models that replicate their unique expertise, allowing their insights to be accessible at any time. This concept could be revolutionary in fields where specialised knowledge is critical, like project delivery, engineering, and construction. Imagine project teams leveraging an AI model trained on the unique skills of a veteran project manager or engineer, making informed decisions even in their absence.

 The article suggests that, as AI advances, building personalised models could allow individuals to scale their expertise and remain influential across multiple projects simultaneously. For industries dependent on high-stakes decisions, this ability to consult a “digital twin” of an expert could streamline decision-making and increase productivity.

However, widespread adoption of personal AI models raises significant questions. Ownership, ethics, and data security become critical, especially if models capture deeply specialised, individual knowledge. As AI technology matures, professionals must ensure that this “self-exporting” serves as a tool to enhance rather than replace the human factor essential to collaborative work. For more, check out the full article here.

💙 Other productivity news we’re loving

• 
  OpenAI partners with media giant Hearst, publisher of Cosmopolitan and Elle, aiming to integrate AI-driven content into traditional media, reflecting the growing intersection of AI and journalism. Read more.

• Zoom introduces AI-powered digital avatars that can speak on users' behalf in meetings, offering a new layer of convenience and presence while raising questions about communication authenticity. Learn more.

• Adobe expands its content authenticity initiative, providing creators with tools to label AI-generated content, aiming to protect original work and ensure transparency in an era of synthetic media. Explore more.

Event of the Week: Project Hack 23 – Sold Out!

Project Hack 23, the highly anticipated 2-day hackathon hosted by Projecting Success, has officially sold out! Taking place at Rolls Royce’s Learning & Development Centre on October 22nd and 23rd, this event gathers enthusiasts and professionals from project management, data analytics, IT, and business analysis fields to tackle real-world challenges with data-driven solutions.

 Project Flux will be reporting live from the event, covering highlights, insights, and the latest in project management innovations. The event promises intense hackathon sessions, workshops, and tutorials on cutting-edge tools, with themes ranging from Net Zero Carbon and Health & Safety to NLP and Artificial Intelligence. It’s a fantastic opportunity for hands-on experience, networking with industry experts, and the chance to win part of the £5k prize pool!

Stay tuned for our on-the-ground coverage as Project Hack 23 pushes the boundaries of data in project delivery.

Podcast of the Week: AI and the Future of BIM with Professor Sarah Davidson

In this episode of Project Flux, we sit down with Professor Sarah Davidson, a leading expert in Building Information Modelling (BIM) and information management. we explore her journey into the field, unpacking the evolution of BIM and the exciting ways AI is enhancing information management. We address the common misconception that BIM is just about 3D modelling—when, in reality, it’s a vital framework for managing data effectively.

Sarah shares insights into how AI can boost decision-making and risk analysis, emphasising that AI should support, not replace, human expertise. We also discuss the importance of clear information specifications and the role of NEMA in guiding the industry. As we look to the future, we touch on real-time APIs, virtual environments, and how integrating various data sources can transform BIM and AI practices.

Key Takeaway: AI and BIM can work together to build smarter, data-driven project environments. Tune in to hear how collaboration, clarity, and a human-centred approach are vital for success.

One More Thing

AI Detectors Fail: USA Founding Fathers Mistaken for ChatGPT Ghostwriters!

Meet Project Flux: About Us

At Project Flux, we're committed to pioneering the future of construction and project delivery through the lens of cutting-edge Artificial Intelligence insights. Our vision is to be at the forefront of integrating AI into the fabric of project delivery, transforming how projects are conceptualised, planned, and executed.

Our Mission Project Flux aims to not only inform and educate but also to inspire professionals in the construction industry to embrace the transformative potential of AI. We believe in the power of AI to revolutionise project delivery, making it more efficient, predictive, and adaptable to the dynamic demands of the modern world.

What We Offer Through our insightful newsletters, podcasts and curated content on LinkedIn, and engaging discussions, Project Flux serves as a resource for professionals seeking to stay ahead in their field. We offer a blend of practical advice, thought leadership, and the latest developments in AI and construction technology.

What People Say…