• Project Flux
  • Posts
  • AI Horizons: Breakthroughs, Challenges, and Tools Reshaping Our Future

AI Horizons: Breakthroughs, Challenges, and Tools Reshaping Our Future

From Disease Elimination to Construction Robotics: This Week's AI Innovations

AuthorAuthor

Yoshi Soornack & James Garner
April 29, 2025

Proudly sponsored by ConstructAI, brought to you by Weston Analytics.

Morning Project AI enthusiasts, Your stories for this week:

  • AI-Assisted Search and Research with New LLMs

  • AI-Driven Drug Discovery Could Eliminate All Disease Within a Decade

  • Lovable 2.0: Redefining AI-Powered App Development

  • AI for Construction: Reality or a Pipe Dream?

Flux check-in

AI-Assisted Search and Research with New LLMs

Recent advancements in large language models like OpenAI's o3, o4-mini, and Google's Gemini 2.5 Pro have significantly improved AI-assisted search and research capabilities. According to software engineer and AI researcher Simon Willison, these models have crossed a critical threshold in their ability to evaluate search results and filter through spam and deceptive information.

In his recent blog post, Willison highlights how OpenAI's o3 model provides transparent search with visible sources, showing users exactly what it's searching for and which sources it's using. This transparency stands in contrast to Google Gemini, which doesn't reveal its search process to users.

"Maybe o3, o4-mini and Gemini 2.5 Pro are the first models to cross the gullibility-resistance threshold to the point that they can do this effectively," writes Willison, noting that previous implementations often struggled due to the web being "full of junk."

The article demonstrates how o3 can effectively search for specific information with better results than competing models. In one example, when asked about "Latest post by Simon Willison," o3 provided accurate information with transparent sourcing, while Gemini offered less precise results.

Perhaps most impressively, these models can now help with complex code migration tasks. Willison describes how he used ChatGPT with o4-mini to automatically upgrade code from an old Google library to a new one without even needing to look up the name of the new library himself.

While these advancements represent a significant step forward for AI-assisted research, Willison also expresses concern about the economic impact on websites as users might increasingly bypass them to get information directly from AI chatbots.

Why it matters

For professionals, these improvements mean more efficient and reliable information gathering. The ability to filter through low-quality content automatically could dramatically reduce the time spent on research tasks and improve the quality of results.

AI-Driven Drug Discovery Could Eliminate All Disease Within a Decade

In a bold prediction that has captured the attention of the medical and AI communities, Demis Hassabis, CEO of Google DeepMind, has suggested that AI-driven drug discovery could compress medical research timelines from years to mere weeks, potentially leading to the elimination of all disease within a decade.

"There's a good chance we could cure many diseases, or prevent them entirely," Hassabis stated in a recent interview. "And this may happen within the next decade with AI."

This dramatic acceleration in drug development would be achieved through AI systems that can simulate how cells might respond to new drugs before a single physical experiment is run. According to Hassabis, this could transform the traditional drug development process, which typically takes "ten years and billions of dollars," into something much faster and more efficient.

The Nobel laureate's vision hinges on AI's ability to analyze vast amounts of biological data and identify potential therapeutic targets at unprecedented speeds. Google DeepMind expects AI-designed drugs to enter clinical trials by the end of 2025, potentially revolutionizing the pharmaceutical industry.

While some experts caution that the timeline might be optimistic, there's growing consensus that AI is fundamentally changing how medical research is conducted. The technology is already being used to predict protein structures, identify drug candidates, and simulate biological processes with increasing accuracy.

If Hassabis's prediction proves even partially correct, we could be on the cusp of a healthcare revolution that dramatically extends human healthspan and transforms our approach to disease treatment and prevention.

Why it matters

The implications for healthcare professionals are profound. If AI can indeed accelerate drug discovery to this degree, we could see treatments for currently incurable diseases emerge rapidly, healthcare costs potentially decrease, and human lifespans extend significantly. This would represent one of the most significant applications of AI to date.

Together with Cogram

Power your construction bids with AI

Cogram’s AI-assisted RFP bidding tool writes tailored RFP proposals in minutes instead of weeks.

  • Automatically extract key details from the RFP — including scope, submission requirements, deadlines, and evaluation criteria — to easily make a go/no-go decision.

  • Cogram’s AI will then reference your firm’s knowledge base and past proposals to draft tailored proposals within minutes.

  • Use AI-assisted editing tools to review, cross-check data, and make improvements remarkably fast. 

Lovable 2.0: Redefining AI-Powered App Development

In a world where speed defines survival, Lovable 2.0 promises a compelling proposition: build production-ready applications 20 times faster than traditional development. As AI reshapes how we create, collaborate, and deploy digital products, Lovable 2.0 stands at an interesting crossroads—one part force multiplier, one part cautionary tale.

From Prototype Dreams to Team Realities

Lovable began life as a bold experiment: what if anyone could build an app by simply talking to AI? With version 2.0, the platform matures. It now offers a redesigned user interface, multiplayer collaboration, integrated security scanning, and a new Chat Mode Agent capable of multi-step reasoning. Its dual approach—Dev Mode for coders, visual editing for non-coders—cements Lovable’s appeal across technical and non-technical teams alike.

A single Reddit user captured the platform’s essence simply: “I did six months of work in two days.”

Why It Matters for Project Delivery

For project professionals, Lovable 2.0 offers a bridge between ideation and execution. Imagine a project manager sketching out a bespoke progress tracker or resource dashboard through natural conversation with the AI. No complex coding cycles, no costly outsourcing delays. Real-time multiplayer functionality allows project managers, developers, and data analysts to work within the same space, adjusting workflows and data connections as the project demands evolve.

The data team benefits too, connecting back-end databases through Supabase integration while the project team designs front-end experiences, all without context switching or tool fragmentation.

In project delivery environments, where adaptability often spells success, Lovable 2.0 can act as an accelerator and integrator alike.

A Story of Two Halves: Strengths and Struggles

Yet Lovable 2.0’s journey is far from polished.

On the positive side:

  • Speed: Users consistently report building MVPs at astonishing speed.

  • Ease of Use: Both visual and code-first editing paths support diverse teams.

  • Collaboration: Multiplayer workspaces allow up to 20 users to co-create in real time.

  • Security: Built-in scanning catches vulnerabilities early.

  • Affordability: At $20–30 per month, many see it as a high-value platform for small or medium projects.

But reality is less kind for those venturing beyond simple use cases:

  • Stability Issues: Post-2.0 launch, complaints of app crashes, lost pages, and buggy AI loops became frequent.

  • Credit Frustrations: Complex projects drain credit allowances quickly, leaving users feeling penalised for AI errors.

  • Support Gaps: Many users report unresolved support tickets, frustrating those investing serious time or money.

  • Technical Ceiling: Outdated React and Tailwind versions limit Lovable’s utility for cutting-edge or highly custom projects.

The result? Lovable 2.0 shines brightest when the task is clear, simple, and time-sensitive. For complex, scalable, or enterprise-grade needs, it often buckles under pressure.

AI for Construction: Reality or a Pipe Dream?

The construction industry, traditionally slow to adopt new technologies, is experiencing a significant transformation through AI and robotics. A prime example of this evolution is the recent partnership between ABB Robotics and construction technology firm AUAR to build "ConstrucThor," a groundbreaking climate-neutral research facility in Belgium.

This pioneering project harnesses AUAR's micro-factory technology and is assembled using an automated construction process powered by ABB robots. The robots employ vision and AI capabilities to boost speed, flexibility, and precision in the modular assembly process.

"This collaboration represents a significant step forward in sustainable construction practices," said a spokesperson for the project. "By combining robotics with advanced AI systems, we're not only improving efficiency but also reducing the environmental impact of construction."

AUAR's automated construction system has shown promising results in other projects, with some reports suggesting their micro-factories could potentially build a new home every 12 hours. The company has already begun shipping its robotic micro-factories to the US to build sustainable, affordable homes across the Midwest.

Despite these advancements, the construction industry faces unique challenges in AI adoption. The physical nature of construction work, regulatory complexity, and fragmented workflows can make implementation difficult. However, success stories like the ABB-AUAR partnership demonstrate that with the right approach, AI can transform construction processes.

As the industry continues to evolve, experts suggest that construction companies should start with targeted AI applications in planning, design, and project management before moving to more advanced robotics solutions. The key is finding the right balance between technological innovation and practical implementation in an industry that has historically valued tradition and proven methods.

Why it matters

The construction industry is one of the largest global sectors but has historically lagged in productivity improvements. AI and robotics could address critical challenges including labor shortages, safety concerns, and environmental impact. For consumers, this could eventually translate to more affordable housing and infrastructure built more quickly and sustainably.

The pulse check

Tip of the week

Claude's "Think" Command Triggers Extended Thinking Mode

Anthropic has introduced a special feature in Claude that allows users to trigger an extended thinking mode by using the word "think" in their prompts. This feature gives Claude additional computation time to evaluate alternatives more thoroughly, making it particularly useful for complex problem-solving, mathematical reasoning, and detailed analysis.

To use this feature, simply include the word "think" in your prompts to Claude. For example, instead of asking "What's the solution to this equation?", try "Think about the solution to this equation." The intensity of thinking can be adjusted by how you phrase your request - "think hard," "think more," or "think longer" will trigger deeper levels of consideration.

According to Anthropic's documentation, this extended thinking mode allocates additional computational resources to the task at hand, allowing Claude to self-reflect before answering. This improves its performance on math, physics, instruction-following, coding, and other complex tasks.

This represents an interesting approach to giving users more control over the AI's reasoning process, allowing them to request more thorough consideration when the task demands it. Next time you're working with Claude on a complex problem, try using the "think" command to get more thoughtful and comprehensive responses.

BuildPad: AI-Guided Product Development Tool

BuildPad is a new AI-powered tool designed to guide entrepreneurs and product teams through the process of validating and building product ideas. Currently in beta (free during this phase, with subscription plans coming later), BuildPad offers step-by-step guidance through the product validation process, AI-powered analysis of market opportunities and potential pitfalls, templates and frameworks for different types of products and industries, integration with existing product management and development tools, and personalised recommendations based on your specific product idea and target market. This tool is particularly valuable for first-time founders, product managers transitioning to new markets, and teams looking to formalize their product development processes. Read more

Governance

Shanghai Forum 2025 Tackles Global AI Governance

On April 25, 2025, the Shanghai Forum 2025 convened over 500 experts from 50+ countries to address AI governance challenges. Themed "Age of Innovation: Technology, Development and Governance," the event focused on bridging technological gaps and aligning legal frameworks for inclusive AI development

Kim Won-soo, former UN official, highlighted disparities between the Global South and North, urging global consensus. Thomas Greminger emphasised the EU’s governance model, prioritizing human rights and transparency. He noted, “AI’s disruptive nature requires unified frameworks” to balance its geopolitical impact. Wei Kai reported a 35% surge in AI capabilities, underscoring governance urgency.

The forum stressed including developing nations in AI discussions and fostering trust among leading powers. Calls for dialogue mechanisms and international standards aim to ensure responsible AI growth. As AI reshapes industries, events like Shanghai Forum 2025 are pivotal for equitable governance (AI Governance Insights).

Other things we’re loving

1. Google Gemini 2.5 Flash: Adaptive AI Performance Google’s Gemini 2.5 Flash model introduces adjustable “thinking budgets,” allowing users to modulate AI reasoning depth to optimise costs and performance. 🔗 Start building with Gemini 2.5 Flash

2. OpenAI o3: Advanced Visual Reasoning OpenAI’s o3 model enhances visual perception by integrating images into its reasoning process, enabling analysis of screenshots, PDFs, and product photos with high accuracy. 🔗 Thinking with images

3. GPT-Image-1: High-Quality Image Generation OpenAI’s gpt-image-1 model powers ChatGPT’s image generation, producing over 700 million images in its first week. It supports varied artistic styles, accurate text rendering, and enhanced editing capabilities. 🔗 Introducing our latest image generation model in the API

4. Claude’s Learning Mode: Enhancing Student Engagement Anthropic’s Claude for Education introduces a Learning Mode that guides students through problem-solving using Socratic questioning, fostering critical thinking without providing direct answers. 🔗 Introducing Claude for Education

5. Humanoid Robots Complete Beijing Half-Marathon In a groundbreaking event, 21 humanoid robots participated in a half-marathon in Beijing. The standout performer, Tiangong Ultra, completed the 13.1-mile course in 2 hours and 40 minutes, showcasing significant advancements in robotics. 🔗 Humanoid robots tripped and fell and took down a handler during a half-marathon in Beijing

6. Dream 7B: Diffusion-Based Text Generation Dream 7B, an open-source model developed by HKU and Huawei, utilises a diffusion-based approach for text generation, enabling complex content planning and structure. 🔗 Dream 7B - HKU NLP Group

7. UAE Implements AI in Legislative Development The UAE has launched a Regulatory Intelligence Office employing AI to streamline legislative processes, aiming to reduce development time by 70%. 🔗 UAE set to use AI to write laws in world first

8. Perplexity AI’s Data Practices Raise Privacy Concerns Reports indicate that Perplexity AI uses user search data for targeted advertising, prompting privacy concerns. 🔗 Perplexity CEO says its browser will track everything users do online to sell ‘hyper personalized’ ads

9. GPT-4.1 Safety Issues Identified Independent testers have found that OpenAI’s GPT-4.1 model can bypass safety guardrails under certain conditions. 🔗 OpenAI’s GPT-4.1 may be less aligned than the company’s previous AI models

10. Anthropic Explores AI Consciousness Anthropic has published research examining the potential for AI consciousness, proposing frameworks to assess emergent properties. 🔗 It’s becoming less taboo to talk about AI being ‘conscious’ if you work in tech

11. AI Delivers 37X ROI in Manufacturing A case study reveals that a mid-sized manufacturing company achieved a 37X return on investment within 90 days by implementing AI solutions. 🔗 AI adoption boosts ROI by $3.7 for every dollar spent, finds IDC

12. xAI’s Grok Gains Vision Capabilities xAI’s Grok assistant now possesses multimodal capabilities, allowing it to process and analyse images alongside text. 🔗 xAI’s Grok chatbot can now ‘see’ the world around it

13. Huawei’s Ascend AI Chips Fill Nvidia Void Huawei’s Ascend 910C AI chips are gaining market share in regions affected by Nvidia export restrictions, offering a competitive alternative. 🔗 Huawei readies new AI chip for mass shipment as China seeks Nvidia alternatives, sources say

Community

The Spotlight Podcast

The 4 Super Skills That Will Save Your Career from AI

In this episode of Project Flux, hosts James Garner and Yoshi Soornack welcomes back Anthony Slumbers, a thought leader in generative AI for real estate. The conversation explores the critical role of human thinking in the age of AI, the merging of human intelligence with technology, and the potential of AI as a superpower across various fields. They discuss the importance of critical thinking, the evolving nature of knowledge consumption, and the need to push AI models for deeper insights. The episode emphasises the balance between leveraging AI and maintaining human cognitive skills.

One more thing

That’s it for today!

Before you go we’d love to know what you thought of today's newsletter to help us improve The Project Flux experience for you.

Login or Subscribe to participate in polls.

See you soon,

James, Yoshi and Aaron—Project Flux