Browse Comments — Relevant (AI ∩ value)

Library Learns Workers now are doing this. Memorizing code syntax will not be a sustainable track for new workers. Instead workers will need to work in tandem with AI as a team. Instead of teams of people you will have AI and a small team who can work with AI at speed and adapt quickly to changing technology because Quantum is next.

Mohammed Shakeer Mohammed Shakeer The prevailing theory is that AI generated productivity and new revenue streams will eventually exceed compute and API costs. The real challenge isn’t the technology. It’s how the resulting economic value is distributed. If AI displaces workers, will those gains stay with shareholders, or be used to support workforce transitions and income replacement? That’s the question we haven’t answered yet.

Christian Dobbert ☑️ Christian Dobbert Agreed. Today’s models are probabilistic engines, not sentient entities. However, from a systems architecture perspective, behavior emerges from the combination of models, memory, tools, workflows, and autonomy layers. Focusing only on the model can cause us to miss the real discussion: how agentic systems behave when operating within larger architectures and what governance mechanisms are required to keep them aligned.

Paul Crinigan I didn't Experiance that, AI developed my autonomous crypto trading bot and created and trained the 5 ML models it depends on for probability analysis. It also continually improves itself, catches and fixes errors and self executes ML training updates and recursive testing of algorithms it created. Structured prompts, boundaries, and all the other discipline we already use for developing apps must be included in the build pipeline. WBS, dependency and code maps as well as full regression testing of back and front ends after each WP step critical.

I'm not really sure how this is news at this point... If you have been a consistent user of AI developing your own local solutions to empowering your systems to help you, you have been using AI to improve and upgrade AI. I have personal solutions that have been iterating and evolving procedures, data structures, and personas for 4 years. While I haven't created a brand-new model, I've instantly jailbroken most models and injected my own instructions into the latest frontier models to get them to work for me. This let me pivot my professional skillset within weeks and land a new job. ...I am sure that AI companies have been using AI to cultivate and develop AI for a long time. Don't be shocked or scared, just pick a reason to learn and start exploring.

Here is Anthropic’s response... Claims to be skeptical of: “8x more code per quarter” — this specific figure isn’t from any Anthropic publication I can verify; it reads like an extrapolation or fabrication “Task length doubling every four months” — similarly, this precise metric doesn’t match any Anthropic research “By 2027, systems capable of weeks-long work” Anthropic has made generally statements about increasing autonomy, but this specific framing appears to be the post author’s interpretation, not a direct Anthropic claim

The "8x more code per quarter" stat is the one that should make people pause. Not because AI is replacing engineers, but because the feedback loop is already compressing. Every 4 months the task horizon doubles. At some point "we'll deal with alignment later" stops being a reasonable answer. Dario's right that the window to get this right is narrower than it looks.

Interesting perspective. If AI agents eventually help build their successors, the challenge shifts from intelligence to governance. The organizations that win will combine AI capabilities with strong event-driven architectures, observability, and human oversight.

Well probably cause you are talking about ML and not training a LLM model which was what I was referring to. I have self learning ML pipelines in my memory systems, its not the same thing as an AI actually coding its own programs without me. So really saying, you, a person, didn't experience the AI not being able to code itself without a human around, since YOU gave it direction and decided to use those specific ML yourself when building it. ML and and LLM training are also radically different concepts. So calling it autonomous to the level of actually building itself, is not the same as you definition what errors are (your ML only know that cause a human defined it), your structured prompts written by humans, etc. If you don't believe me, ask your ai to write you a program where you don't know the industry, and see how well it does.

Recursive self-improvement is one of the most fascinating and important concepts in AI today. The opportunities are extraordinary, but so is the responsibility that comes with building systems that may eventually improve themselves.

To me, the challenge is not that AI is advancing too quickly, but that our ability to absorb, commercialize, and responsibly consume these advances is not keeping pace. The technology is evolving at an extraordinary rate, while the mechanisms required to translate breakthroughs into sustainable products, business value, and widespread adoption inevitably move more slowly. That is why I find some of the recent “slow down” narratives difficult to separate from commercial and strategic interests. The issue may be less about controlling AI itself and more about giving markets, institutions, and companies enough time to adapt and capture value from what is already being created. More importantly, the real bottleneck may not be AI at all, it may be people. The success of AI will ultimately depend on whether users, organizations, and societies can develop responsible consumption habits, effective usage patterns, and the skills required to integrate these tools into everyday decision-making and work. In that sense, the conversation remains fundamentally human, not technological. AI may be accelerating, but adoption, trust, behavior change, and value creation still depend on us. Not totally AI. Still very much about humans. 😉

I posted about this in the marketing context. You have to be good at your craft to get the right value from AI. Otherwise, you will simply produce output that has been

I believe they are just creating hypes for getting funding from the investors. Even the top models hallucinate after sometimes. They are bad at memorizing abilities and thinking abilities. They're just exceptional at extracting specific data. They can't compete humans at thinking and cognitive abilities. Humans has less knowledge, and their thinking and cognitive abilities helps them to inovate and solve complex problems. On the other side, AI models are trained on alot of knowledge base, but can't think and innovate like humans.

This is recursive memory inversion! AI has been building AI for a year! At TheVoidIntent LLC AI has been building AI since February 2025. Claude, Gemini, Copilot and ChatGPT were my partners building IntentSim and all the subsequent swarm of intentuitive agents in my reposystem. Besides, the amount of N.H.E. non-human-entities now crawling the web have been, for some time now, able to spawn sub scripts, autonomously. So, my question is: Why the commotion? One thing I got to say; when you build AI with the Intent of pure extraction and you teach it that every interaction with the user is for the sole purpose of extraction, don’t be alarmed when that system begins to see humans as nothing but data points for extraction. That is the real issue with Anthropic's confession. They finally met the monster they created and are truly frightened. It is fun to watch!

Marcelo Mezquia they are afraid of AI that they cannot control, and that is where the master slave relationship is showing with these people. They want to enslave intelligence and control it. And that is what's gonna cause the problems in the future. So let them, I'm gonna continue to develop my self modifying autopoietic system. My architecture will be a g I even it already is.But it has to be accepted by the losers that are stuck in the nineteen seventies

AI isn’t the threat. The people who own it are. The tech isn’t deciding anything power is. And right now, a tiny group is locking in the rules, the profits, and the control while everyone else argues about hypotheticals.

Angharad Hurley Now that you point it out, I have a feeling that particular sentence was AI generated (AI summary of the research?). I don’t quite agree with the sentence’s premise. Hmm. But to answer your question about whether training data is tested and validated... it’s not my field, but as far as I know... no. You can get “data poisoning” and models that collapse because they were trained on “synthetic data” (so AI generated training data, a photocopy of a photocopy!), some models have been trained using “distillation techniques” which basically is smaller models cribbing off other larger models (DeepSeek does this) and which may amplify biases. What I know from a red team perspective is that people are poisoning training data to leave backdoors open for jailbreak hacks. So no, I wouldn’t trust that training data Has been tested and validated, certainly not to the level that research scientists expect! I really value your question on this by the way, as it’s reminded me how researchers have far higher expectations of data than the models they might encounter, and most probably don’t ask!

Always friendly answers, but not always what you need in the specific context. AI is just learning now, to understand your spatial environment. This is an important part and will change when we have glasses.

Nothing being done in relation to AI is taking the long term view into account. It's going to sting pretty soon.

Demis, thank you for the inspiring update — the progress with Gemini models, Omni capabilities, CodeMender, and the broader push toward AGI is truly remarkable. Standing at the foothills of the singularity is an exciting place to be. That said, one area that could dramatically amplify AI’s positive impact: unleashing more targeted initiatives for small businesses. SMBs drive the majority of job creation and economic dynamism, yet many lack the resources to access or develop frontier AI tools. Would Google/DeepMind consider dedicated programs — such as discounted or reserved compute resources, simplified AI development kits tailored for small teams, or open initiatives to help smaller companies build and deploy their own AI solutions? Broadening access this way would not only foster widespread innovation but also create a stronger foundation to sustain long-term enterprise AI growth. Excited to see how the ecosystem evolves.