Ideas Worth Exploring: 2025-04-24
- Charles Ray
- Apr 24
- 6 min read
Ideas: Andrew Chow - AI Outsmarts Virus Experts in the Lab, Raising Biohazard Fears
A recent study reveals that advanced AI models like ChatGPT and Claude now outperform PhD-level virologists in solving complex lab problems, raising both significant benefits and concerns in the field of biomedicine. The study, conducted by researchers from multiple institutions, tested these models' ability to troubleshoot intricate lab procedures and protocols. The results showed that while human experts scored around 22%, AI models scored up to 43% accuracy.
This development has a double-edged impact: it can accelerate disease prevention and cure discovery, as suggested by OpenAI's CEO and confirmed by experts like Tom Inglesby from Johns Hopkins Center for Health Security. However, it also raises serious security concerns. Non-experts could potentially weaponize these models to create ideas for deadly bioweapons, as they now have access to a non-judgmental AI virology expert capable of walking them through complex lab processes.
GitHub Repos: c/ua - Computer-Use Agent
c/ua (pronounced "koo-ah", short for Computer-Use Agent) is a framework that enables AI agents to control full operating systems within high-performance, lightweight virtual containers. It delivers up to 97% native speed on Apple Silicon and works with any vision language models.
c/ua offers two primary capabilities in a single integrated framework:
High-Performance Virtualization - Create and run macOS/Linux virtual machines on Apple Silicon with near-native performance (up to 97% of native speed) using the Lume CLI with Apple's Virtualization.Framework.
Computer-Use Interface & Agent - A framework that allows AI systems to observe and control these virtual environments - interacting with applications, browsing the web, writing code, and performing complex workflows.
Ideas: Jess Brown - Practical ways I use AI in design
Jess Brown a product designer at Faire, shares experiences integrating AI into their design practice in early 2025. They explore two main applications: informing design decisions and generating content.
Informing Design Decisions:
The designer uses custom chatbots to quickly access internal knowledge, like support tickets and policies, aiding understanding of customer pain points and complex program details.
AI language models (LLMs) are employed to synthesize interview transcripts, organizing insights efficiently and enhancing cross-interview analysis. This helps identify patterns, such as customer confusions or delights.
AI assists in wrangling large datasets, like filtering and structuring trade show data, making it easier to navigate events and connect with relevant customers.
LLMs can also enrich data by filling in missing information or combining mismatched formats, creating comprehensive views without manual effort.
Generating Content:
AI helps brainstorm microcopy, expanding idea range and refining phrasing for conciseness and brand alignment.
The designer uses AI to speed up light market research, gathering competitive go-to-market messages quickly.
LLMs generate realistic mock data, such as retailer names and product names, making high-fidelity mockups more representative.
Ideas: Scott Berkun - Rethinking bad design
Scott Berkun discusses the prevalence of poorly designed products despite living in an era of advanced technology. Scott Berkun argues that this is not due to user incompetence, but rather flawed design choices made by corporations. Scott Berkun highlights several examples, including vaccine registration websites, banking software, and trading apps, which have caused significant harm due to poor usability.
The root of the problem lies in the history of design itself. Decades ago, researchers like Paul Fitts proved that aircraft crashes were often due to cockpit design, not pilot error. However, this insight is frequently overlooked, as seen in projects like the F-35 and Quibi, which failed due to poor design decisions. Scott Berkun attributes this to several factors: technology and design being taught separately, homogeneous engineering teams lacking diverse perspectives, and businesses prioritizing sales over usability.
To address these issues, Scott Berkun proposes shifting our focus from technological potential to actual user achievements, valuing design expertise, and educating consumers, technologists, and business leaders about the importance of good design. By doing so, we can hold corporations accountable for producing usable, safe, and accessible products that serve the needs of all users.
Ideas: Ryan Whitwam - Google won’t ditch third-party cookies in Chrome after all
Google has unexpectedly abandoned its long-standing plan to phase out third-party cookies in its Chrome browser, a project initiated under the banner of the Privacy Sandbox. After years of experimenting with different alternatives like FLoC (Federated Learning of Cohorts) and the Topics API, Google has decided not to implement a feature that would have prompted users to disable third-party cookies. Instead, cookie support will remain unchanged, potentially indefinitely.
The decision comes amidst criticism of Google's earlier proposals and repeated delays in its plans to deprecate cookies. While Google maintains that it remains committed to preserving user privacy, the advertising industry's increased focus on privacy seems to have influenced its choice to maintain the status quo. Users can still manually disable third-party cookies in Chrome if they choose to do so.
Ideas: Ernestas Naprys - Massive botnet bigger than some countries discovered as DDoS attacks soar
The cybersecurity landscape is facing a significant threat due to the increasing size and number of botnets composed of end-of-life devices connected to fast internet. A record-breaking botnet in 2025 comprised of 1.33 million devices, surpassing the previous year's largest botnet by six times, underscores this alarming trend. This massive botnet, if considered a country, would rank among the top 60 globally in terms of population.
The growth and power of such botnets are attributed to the proliferation of cheap, outdated, and vulnerable devices, particularly in developing countries where economic constraints lead people to keep devices long after their support ends. These devices often lack security updates and can be easily compromised, turning them into "zombie" devices controlled by hackers.
The consequences are evident: a 110% increase in DDoS (distributed denial of service) attacks compared to the previous year. These attacks primarily target industries like IT and telecoms, fintech, and e-commerce, with the potential to generate tens of millions of requests per second, overwhelming unprotected systems.
Experts warn that while geo-blocking can be an initial mitigation strategy, bot operators are prepared to switch to IP addresses from other regions, making such attacks increasingly difficult to thwart. As the trend of massive DDoS botnets continues to rise, it underscores the urgent need for improved device security and disposal practices in developing countries.
Ideas: rosie - Ancient History, Modern Failures: How Crypto Projects Follow a 2,200-Year-Old Script
In an insightful piece, the author draws parallels between crypto governance and Polybius's political revolution cycle, highlighting a predictable trajectory towards recentralization. The journey begins with a "Visionary Founder" phase, where developers, driven by a manifesto to liberate society from central authority, assert their power as moral authorities, making crucial decisions while professing community equality.
As the project grows, this founder's influence becomes problematic, ushering in the "Cult of Personality" phase. Decision-making becomes self-serving, criticism is silenced, and the project increasingly reflects the leader's ego rather than user needs. This is followed by the "Core Dev Priesthood" phase, where a small group of core developers, supposedly chosen based on technical merit, takes control. Wealth concentration eventually leads to an "Ecosystem Fund" phase, with large token holders forming an unofficial ruling class that decides outcomes privately.
Community rebellion against whale control then initiates the "DAO Revolution," with governance tokens distributed more widely and participation initially high. However, this enthusiasm fades, leading to the "Twitter Tantrum" phase, where mob rule prevails, and decisions cater to loud voices rather than long-term thinking or merit. The cycle completes its loop as projects return to centralized control under the guise of restoring order.
The author concludes that each revolution around this carousel results in abandoned Discord servers, worthless tokens, and unlearned lessons. To break this cycle, projects must acknowledge that perfect decentralization leads to poor governance, instead finding a balance between decentralization for security and coordination for effective decision-making. Until then, the cycle continues, with Polybius profiting from trading governance tokens based on his predictive understanding of revolutionary trends.draws parallels between crypto governance and Polybius's political revolution cycle, highlighting a predictable trajectory towards recentralization. The journey begins with a "Visionary Founder" phase, where developers, driven by a manifesto to liberate society from central authority, assert their power as moral authorities, making crucial decisions while professing community equality.
As the project grows, this founder's influence becomes problematic, ushering in the "Cult of Personality" phase. Decision-making becomes self-serving, criticism is silenced, and the project increasingly reflects the leader's ego rather than user needs. This is followed by the "Core Dev Priesthood" phase, where a small group of core developers, supposedly chosen based on technical merit, takes control. Wealth concentration eventually leads to an "Ecosystem Fund" phase, with large token holders forming an unofficial ruling class that decides outcomes privately.
Community rebellion against whale control then initiates the "DAO Revolution", with governance tokens distributed more widely and participation initially high. However, this enthusiasm fades, leading to the "Twitter Tantrum" phase, where mob rule prevails, and decisions cater to loud voices rather than long-term thinking or merit. The cycle completes its loop as projects return to centralized control under the guise of restoring order.
The author concludes that each revolution around this carousel results in abandoned Discord servers, worthless tokens, and unlearned lessons. To break this cycle, projects must acknowledge that perfect decentralization leads to poor governance, instead finding a balance between decentralization for security and coordination for effective decision-making. Until then, the cycle continues, with Polybius profiting from trading governance tokens based on his predictive understanding of revolutionary trends.
Comments