- TechX Insides
- Posts
- 🔒 AT&T Leak, ⚙️ SLMs vs LLMs, & 🧠How to Make Better Decisions
🔒 AT&T Leak, ⚙️ SLMs vs LLMs, & 🧠How to Make Better Decisions
AT&T data leak, Microsoft Stargate, SLMs vs LLMs, OpenAI Voice Clone; and more
Your Tech World Pocketbook: News, Lingo, Learning
Tech News & Trends 📢
AT&T resets millions of customer passcodes after a massive leak of sensitive data was made public in March 2024
AT&T has reset millions of customer account passcodes following a significant data leak from 2019, impacting 7.6 million current and 65.4 million former account holders. The leaked data - which includes names, addresses, Social Security numbers, and encrypted account passcodes - was made public earlier this month. AT&T is investigating with cybersecurity experts but has found no evidence of unauthorized system access. The source of the leak, whether from AT&T or a vendor, remains unclear.
LINK: Techcrunch (3 min)
OpenAI claims that it can clone a voice from just 15 seconds of audio, and produce natural-sounding speech
OpenAI has developed Voice Engine, a tool capable of cloning any voice from just a 15-second audio sample, producing natural-sounding and emotive speech. This technology has applications in reading assistance, language translation, and aiding speech-impaired individuals. Concerns about deepfake misuse and privacy are being addressed with safety measures like watermarking and a "no-go voice list" to prevent impersonation of prominent figures. Pricing and further details are under wraps, with potential affordability highlighted.
LINK: Engadget (3 min)
What are Small Language Models (SLMs) and how do they compare against LLMs?
Small language models (SLMs) are carving out a niche for efficiency and task-specific performance. Contrary to large language models (LLMs), SLMs utilize fewer parameters, enabling them to operate on devices with minimal computing power (e.g. smartphones). SLMs excel in applications like sentiment analysis, text summarization, and code generation. Examples like Google’s Gemini Nano and Microsoft's Orca models illustrate their practical applications. Also, these compact models are not only cost-effective but also energy-efficient, promising a sustainable alternative for AI deployment in various domains.
LINK: Indian Express (5 min)
US government directs all federal agencies to hire Chief Artificial Intelligence Officers to oversee adoption and risk mitigation
The White House has mandated all federal agencies to appoint Chief AI Officers. This policy aims to modernize the government's approach to AI, ensuring responsible use and managing risks. Agencies must establish AI governance boards, implement safeguards for AI applications affecting rights or safety, and publish annual lists of AI systems with risk assessments. There are also plans to hire 100 AI professionals across agencies.
LINK: CBS (3 min)
<And More>
5. Microsoft is reportedly building Stargate, a $100b supercomputer, to power the next generation of OpenAI models
LINK: Gizmodo (5 min)
6. Sam Bankman-Fried (SBF), founder of crypto exchange FTX, sentenced to 25 years in prison as a jury found him guilty for fraud and conspiracy
LINK: Reuters (7 min)
7. US and Europe officials alarmed by cheap green tech from China, stoking fears of unfair competition hurting their domestic interests
LINK: Washington Post (5 min)
8. Spacetech investments in India are heating up, alongside strong public sector support with 0% tax and increased government budget allocation
LINK: Inc (7 min)
Dinner Table Lingo đź’¬
Barbell Distribution
A barbell distribution has two peaks at the ends of the spectrum, with fewer values in the middle. This term is metaphorically used in the tech industry and investment strategies to describe a situation where resources, efforts, or investments are heavily concentrated in two distinct areas, often with little in between.
“Considering our current project allocation, we're essentially looking at a barbell distribution. We're heavily invested in both cutting-edge, high-risk technologies and stable, low-risk core products. This approach has minimized our middle-ground investments, reflecting our strategy to innovate while maintaining a reliable base.”
Event Storming
Event Storming is an interactive, group modeling technique used in software development to quickly find and explore complex business domains. It involves gathering domain experts and development team members to brainstorm and visually map out all the events that occur within a system or process.
“By mapping out every event in the logistics process (from pickup to delivery), everyone gained a comprehensive view of the system. This not only clarified the technical challenges but also surfaced new ideas for features that could significantly enhance efficiency. It's a powerful method for rapidly converging on a shared understanding of the project's scope and priorities.”
Further Learning đź’ˇ
How To Make Better Decisions
Here are 3 mental models to improve your critical thinking and decision-making.
👇 1. Inversion
Conventional wisdom = plan well and consistently to achieve your goals.
But what if the way upward/forward to our goals = think in reverse/upside down? Inversion helps surface one’s assumptions (as well as potential uncontrollable factors), and provides clarity on what not to do. Instead of thinking “what do I need to do for this to succeed”, consider “what needs to happen for this to fail”
E.g. NOT “these are the steps required to hit my goal” BUT “how can I fail to hit my goal - and what do I need to stop doing or prevent from happening”?
🔎 2. Framing
A surgeon relays the same information to 2 patients but in different ways
Patient A: You have a 65% chance of living
Patient B: You have a 35% of dying
Patient A opts in for the surgery, while B opts out.
Framing allows you to tap into bias and inclinations (e.g. loss avoidance). Build your understanding of common biases, and shift your communication + planning accordingly. Aim to incorporate urgency and loss avoidance into discussions: e.g. NOT “we need to do X to gain Y” BUT “if we don’t do X quickly to gain Y, we risk losing Z”.
đź–– 3. Bimodal Thinking
Bimodal thinking in organisations involves two modes of working: one focusing on stability and the other on quick actions. This is relevant across many dimensions (e.g. leadership, personal life) - at its core, this means holding two conflicting ideas/approaches simultaneously. Let’s apply this concept to ways of working across your leadership style and broader team management.
Leadership Style: Conduct small experiments & take risks to progress further, BUT ALSO adhere to Standard Operating Procedures (SOPs) to ensure stability
Organisational management: Agile teams sustain innovation, hierarchical siloes retain control. Consider cross-functional brainstorming and projects to enhance an exchange of ideas and execution, BUT ALSO retain current org structure.