Bilingual

EnshittifAIcation
a day ago
- An AI customer support from a digital marketplace demanded disabling HTTP/2, ignored technical explanations, and threatened suspension.
- Increasing botnet attacks on e-commerce servers led to geo-blocking, but a partner failed to provide connection details and insisted on whitelisting their user agent despite firewall-level blocking.
- A marketing consultancy used AI to recommend migrating to an 8GB RAM VPS for a server with 128GB RAM, demonstrating overconfident and incorrect advice.
- Unlike human interns who learn and grow, AI systems cannot update their understanding or admit ignorance, causing confusion and wasted time.
- Companies over-relying on AI for customer service and technical decisions risk more harm than savings, as AI is a stochastic machine with inherent limitations.
- The author questions why AI is trusted over senior professionals but not used to replace bosses, highlighting a human bias toward rewarding confidence over competence.
Notes from Bryan Cantrill’s “Intelligence is not Enough”
a day ago
- Bryan Cantrill discusses company-destroying bugs that had no precedent and were unsolvable by existing documentation or AI reasoning.
- Breakthroughs required human values like resilience, teamwork, rigor, and optimism, not just intelligence.
- Cantrill criticizes the overemphasis on intelligence, advocating for super collaboration and super teamwork instead.
- In hiring, values such as teamwork and optimism are more important than raw intelligence.
- The talk argues that pretending AI can engineer autonomously does a disservice to human humanity.
<antirez>
a day ago
- The author, an experienced programmer, uses LLMs like ChatGPT to accelerate coding by reducing time spent on documentation, learning complex APIs, and creating disposable programs.
- LLMs are often described as a 'stupid but all-knowing savant'; they excel at providing knowledge but struggle with complex reasoning, making them most useful for high-level tasks in Python rather than system programming in C.
- Examples include using GPT4 to write Python scripts for data analysis, convert between machine learning frameworks (e.g., Keras to PyTorch), and create quick utilities like CSV report analyzers.
- LLMs fail at tasks requiring deep reasoning, such as implementing a Bloom filter correctly or reverse-encoding quantization formats, often producing flawed code that requires expert intervention.
- The author argues that LLMs are valuable for modern programming which often involves repetitive code (e.g., API boilerplate) rather than novel algorithms, but warns that their current limitations mean they cannot assist in system programming effectively.
- The article concludes that using LLMs is a practical skill that improves problem description abilities and saves time, but their benefits are skewed toward those already skilled in programming.
No, an AI cannot know the future and never will..
10 days ago
- AI cannot know the future because it relies on human-captured event chains which are incomplete, high-resolution, finite, and limited.
- Human modalities (like text, images) only capture some event chains, often with differing perspectives, resolutions, and arbitrary endpoints, making AI predictions flawed.
- Future AI might try to capture all event chains with a reality sensor array, but faces challenges like the cold start problem, infinite universe size, and self-prediction recursion.
- If AI succeeds in predicting the future by merging with reality, it ceases to exist, highlighting that knowing the future is equivalent to death.
- In a creative context, the author urges breaking free from self-judgment, comparing with idols, and creating original content without paralysis by analysis.
Stop Telling Me to Ask an LLM
18 days ago
- The author prefers consulting experienced individuals over relying solely on LLMs for complex, nuanced questions.
- When seeking expert advice, responses often redirect to asking Claude, even after thorough LLM research.
- Personal experience and subjective judgment are valued over generic or consensus-based information.
- "Ask Claude" may be a polite way to decline or avoid thoughtful engagement, lacking lived insights.
- While LLMs are useful for many queries, they fall short for questions requiring deep, experiential knowledge.
Regression to the Mean: on LLMs and the quiet death of the new
23 days ago
- The promise of AI as a tool for innovation may lead to a flattening of ideas, mistaking uniformity for progress.
- AI models, trained on existing data, generate probable continuations based on past thought, not truth, correcting novelty as error.
- The consensus-building nature of AI can narrow cultural variance, devaluing average outputs and making deviations priceless.
The One-Person Company Fantasy
a month ago
- AI alone cannot replace human judgment, responsibility, and ethical decision-making, making a one-person company improbable.
- Companies rely on a system of complex decisions across areas like technology, operations, and morality, which AI cannot fully navigate without human context.
- AI amplifies human competence but also incompetence, as seen in 'vibe-coded' projects that become hard to maintain, highlighting the need for skilled teams.
- The value of a team lies in shared judgment, accountability, and contextual understanding, not just speed, which AI cannot replicate.

Hasty Briefsbeta

ai limitations

EnshittifAIcation

Notes from Bryan Cantrill’s “Intelligence is not Enough”

<antirez>

No, an AI cannot know the future and never will..

Stop Telling Me to Ask an LLM

Regression to the Mean: on LLMs and the quiet death of the new

The One-Person Company Fantasy