So I guess everyone uses LLMs to think now
eva.ac ·
Eva, a security researcher and blogger, says she has never used AI for research published on her site and prefers manual coding, debugging, reverse engineering, source-code exploration, and vulnerability research. She says LLM output can be confidently wrong and frustrating to supervise, while acknowledging that the models are powerful. Eva argues that current LLMs still require skilled human guidance; she says an unguided model would probably not find a Linux kernel remote-code-execution vulnerability on its own.
Eva criticizes AI companies’ cybersecurity marketing, citing differing accounts of the php2shell research collaboration involving Calif, OpenAI, and Wiz. She says OpenAI emphasized an agent uncovering critical vulnerabilities, while Calif described human-guided AI research. Eva also argues that publishing proof-of-concept videos, product-linked hashes, or cryptic vulnerability announcements before patches are available can create unnecessary risk and panic. She cites a publicly disclosed Tailscale SSH privilege-escalation issue that she says required existing ACL access to a machine with Tailscale SSH enabled, and alleges it was not responsibly disclosed to Tailscale before publication. Eva contrasts AI-assisted findings with the eMudhra certificate-authority private-key incident, which she says a human discovered using Google’s paranoid_crypto library. She argues that personal projects provide learning and enjoyment that AI-generated work can reduce.
What people are saying
The thread spent far less time on the article’s styling than on whether using an LLM means surrendering the satisfying, valuable parts of technical work.
What readers liked
- Modern coding agents can turn sparse documentation into a complete pull request and accompanying explanation, making previously tedious project…
- LLMs are proving valuable when they accelerate research and line-level implementation while a human still owns the goal, review, and technical…
Concerns raised
- A model that confidently follows a foolish path can be intolerable to watch and expensive to run, especially for users paying directly for…
- Longer agent runs may improve apparent results by letting models repair themselves, but they also raise cost, power, and latency instead of…
Read full discussion analysis
Source: Hacker News · Based on 13 sampled comments · Analyzed
Discussion overview
The thread spent far less time on the article’s styling than on whether using an LLM means surrendering the satisfying, valuable parts of technical work. Several commenters reject the premise that refusing AI is inherently principled: if writing software or research is a hobby, doing it manually can be the point, much like playing chess despite engines. But in paid engineering, they argue, tools that draft code, research, or pull requests can save real time, and abstention may be a luxury workers cannot afford.
The sharper concern is not tool use itself but where people draw the boundary. Commenters have seen LLMs make capable developers dramatically faster when used for research, repetitive implementation, and first-pass coding. They have also seen teams hand over judgment, architecture, and whole codebases, then create fragile software and push the verification burden onto reviewers. Skeptics say the models remain maddeningly willing to pursue wrong paths with apparent certainty, often consuming extra tokens and time before self-correcting. Defenders counter that many critics are judging obsolete or free-tier systems, while newer agentic models can complete substantial tasks that were recently unrealistic.
There is little patience for purity politics on either side. The practical consensus is that AI is already changing career expectations, but its usefulness depends on keeping humans responsible for specifications, taste, and final judgment.
Points of disagreement
The central fight is whether modern LLM skepticism reflects outdated experience or a persistent defect in the technology. Optimists say current premium models and long-running agents reliably deliver major productivity gains; skeptics say more elaborate reasoning and tool use often merely hide hallucination behind longer, costlier detours, leaving humans to catch the damage.
What readers liked
- Modern coding agents can turn sparse documentation into a complete pull request and accompanying explanation, making previously tedious project onboarding far faster.
- LLMs are proving valuable when they accelerate research and line-level implementation while a human still owns the goal, review, and technical decisions.
- Delegating routine coding can free developers to work asynchronously or attend to other tasks, a concrete productivity gain rather than a philosophical one.
- Even strong AI-assisted apps can outperform incumbent alternatives with surprisingly little manual implementation, showing how quickly small teams can ship useful software.
Concerns raised
- A model that confidently follows a foolish path can be intolerable to watch and expensive to run, especially for users paying directly for token-heavy retries.
- Longer agent runs may improve apparent results by letting models repair themselves, but they also raise cost, power, and latency instead of eliminating the underlying error problem.
- When people outsource judgment rather than clerical work, reviewers inherit the real labor: detecting bad assumptions, correcting architecture, and cleaning up brittle generated code.
- Workers feel pressured to adopt LLMs to remain employable, turning a supposedly optional productivity tool into an exhausting career requirement.
Keep reading
Sources and details
eva.ac · Published
So I guess everyone uses LLMs to think now
Read report →
Community analysis available for 1 of 1 linked discussions.
Additional discussion searches are still incomplete.
Coverage details and dates
Latest available community analysis:
Summaries: 1 current · 0 awaiting update · 0 not verified
Latest summary update:
Hacker News · eva.ac