HNHacker News
TopNewBestAskShowJobs

nopinsight

7,312 karma · joined March 14, 2009

Co-Founder of a stealth startup

Working on:

agentic systems, natural language understanding, machine learning

Comments here do not represent any companies or organizations I am associated with.

Email: My username at the best known email service beginning with g.

submissionscomments
nopinsight··on Claude Opus 4.6
Some of Opus 4.6's standout results for me:

* GDPVal Elo: 1606 vs. GPT-5.2's 1462. OpenAI reported that GPT-5.2 has a 70.9% win-or-tie rate against human professionals. (https://openai.com/index/gdpval/) Based on Elo math, we can estimate Opus 4.6's win-or-tie rate against human pros at 85–88%.

* OSWorld: 72.7%, matching human performance at ~72.4% (https://os-world.github.io/). Since the human subjects were CS students and professionals, they were likely at least as competent as the average knowledge worker. The original OSWorld benchmark is somewhat noisy, but even if the model remains somewhat inferior to humans, it is only a matter of time before it catches up or surpasses them.

* BrowseComp: At 84%, it is approaching human intersubject agreement of ~86% (https://openai.com/index/browsecomp/).

Taken together, this suggests that digital knowledge work will be transformed quite soon, possibly drastically if agent reliability improves beyond a certain threshold.

nopinsight··on GPT-5.3-Codex
From Claude 4.6 Thinking:

OSWorld is the full 369-task benchmark. OSWorld Verified is a ~200-task subset where humans have confirmed the eval scripts reliably score success/failure — the full set has some noisy grading where correct actions can still get marked wrong.

Scores on Verified tend to run higher, so they're not directly comparable.

nopinsight··on Rentahuman – The Meatspace Layer for AI
Extrapolate a bit to when AI is capable of long-term, complex planning, and you see why AI alignment and security are valid concerns, despite the cynicism we often see regarding the topic.
nopinsight··on Horses: AI progress is steady. Human equivalence is sudden
That depends on the social contract we collectively decide (in a democracy at least). Many possibilities will emerge and people need to be aware and adapt much faster than most times in history.
nopinsight··on Horses: AI progress is steady. Human equivalence is sudden
AI agents might be able to automate 80% of certain jobs in a few years but that would make the remaining 20% far more valuable. The challenge is to help people rapidly retrain for new roles.

Humans will continue to have certain desires far outstripping the supply we have for a long time to come.

We still don’t have cures for all diseases, personal robot chefs & maids, and an ideal house for everyone, for example. Not all have the time to socialize as much as they wish with their family and friends.

There will continue to be work for humans as long as humans provide value & deep connections beyond what automation can. The jobs could themselves become more desirable with machines automating the boring and dangerous parts, leaving humans to form deeper connections and be creatively human.

The transition period can be painful. There should be sufficient preparation and support to minimize the suffering.

Workers will need to have access to affordable and effective methods to retrain for new roles that will emerge.

“soft” skills such as empathetic communication and tact could surge in value.

nopinsight··on A new Google model is nearly perfect on automated handwriting recognition
The reason is given here: https://news.ycombinator.com/item?id=45937772
nopinsight··on A new Google model is nearly perfect on automated handwriting recognition
My comment above serves as a covert commentary on the utility of current frontier LLMs, which imo can often generate higher-quality responses than some HN comments. (And yes, I did agree with their responses above.)

I enjoy the recursiveness of it all. Perhaps I should have said it outright.

nopinsight··on A new Google model is nearly perfect on automated handwriting recognition
The comment above seems to violate several HN guidelines. Curious, I asked GPT and Gemini which ones stood out. Both replied with the same top three:

https://news.ycombinator.com/newsguidelines.html

They are:

1. “Be kind. Don't be snarky. … Edit out swipes.”

2. “Please don't sneer, including at the rest of the community.”

3. “Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something.”

nopinsight··on Why "everyone dies" gets AGI all wrong
What’s your objective ‘threshold’ or a set of capabilities that would compel you to accept a mind as an AGI?
nopinsight··on A definition of AGI
Creative problem solving and commonsense physics are missing, among others.

It is a valuable contribution but the CHC theory from psychology that this is based on is itself incomplete.

By commonsense physics, I mean something like simulating interactions of living and non-living entities in 3D over time. Seems more complicated than the examples in the web site and in most tests used in psychometrics.

Creative problem solving with cognitive leaps required for truly novel research & invention could lie outside the rubrics as well. The criteria in CHC are essential but incomplete I believe.

nopinsight··on AI assistants misrepresent news content 45% of the time
Agreed that current LLMs have low floors despite decently high ceilings.

Note that people who write academic papers are quite far from the median white-collar worker.

nopinsight··on AI assistants misrepresent news content 45% of the time
Hallucination Leaderboard "This evaluates how often an LLM introduces hallucinations when summarizing a document."

https://github.com/vectara/hallucination-leaderboard

If the figures on this leaderboard are to be trusted, many frontier and near-frontier models are already better than the median white-collar worker in this aspect.

Note: The leaderboard doesn't cover tool calling, to be clear.

nopinsight··on Andrej Karpathy – It will take a decade to work through the issues with agents
A definition of AGI: https://www.agidefinition.ai/

A new contribution by quite a few prominent authors. One of the better efforts at defining AGI *objectively*, rather than through indirect measures like economic impact.

I believe it is incomplete because the psychological theory it is based on is incomplete. It is definitely worth discussing though.

—-

In particular, creative problem solving in the strong sense, ie the ability to make cognitive leaps, and deep understanding of complex real-world physics such as the interactions between animate and inanimate entities are missing from this definition, among others.

nopinsight··on Sora 2
Re: the clips above

Although we can tell they are inaccurate, what percentage of people can visualize the prompts better in their mind’s eyes? I bet a substantial number can’t even tell the clips are generated if posted without context.

In a few aspects, these world models are already pretty close to what we have in our brains.

nopinsight··on Sora 2
OpenAI launches Sora 2 in a consumer app to collect RL feedback en masse and improve their world models further.

Their ultimate goal is physical AGI, although it wouldn’t hurt them if the social network takes off as well.

nopinsight··on AI Is Slowing Down Tracker
More than half of the 2024 links, about 15, appeared between o1-preview’s September launch and a few days after o3’s late-December announcement. That span was arguably the most rapid period of advancement for these models in recent years.
nopinsight··on Blue-collar jobs are gaining popularity as AI threatens office work
Humans being who they are, there's still a tremendous amount of work to do in this world (and beyond).

Does everyone alive already have the best quality of life imaginable, not to mention future generations?

Lump of Labor fallacy: https://en.wikipedia.org/wiki/Lump_of_labour_fallacy

Comparative Advantage: https://en.wikipedia.org/wiki/Comparative_advantage

*The key challenge* we all share is making the transition as smooth as possible for everyone involved.

nopinsight··on Curious about the training data of OpenAI's new GPT-OSS models? I was too
The author might use it as an analogy to mentalese but for neural networks.

https://en.wiktionary.org/wiki/mentalese

EDIT: After reading the original thread in more detail, I think some of the sibling comments are more accurate. In this case, neuralese is more like language of communication expressed by neural networks, rather than its internal representation.

nopinsight··on GPT-5
There are quite a few relatively objective criteria in the real world: real estate holdings, money and material possessions, power to influence people and events, etc.

The complexity of achieving those might result in the "Centaur Era", when humans+computers are superior to either alone, lasting longer than the Centaur chess era, which spanned only 1-2 decades before engines like Stockfish made humans superfluous.

However, in well-defined domains, like medical diagnostics, it seems reasoning models alone are already superior to primary care physicians, according to at least 6 studies.

Ref: When Doctors With A.I. Are Outperformed by A.I. Alone by Dr. Eric Topol https://substack.com/@erictopol/p-156304196

nopinsight··on GPT-5
1. FWIW, I watched clips from several of Dario’s interviews. His expressions and body language convey sincere concerns.

2. Commoditization can be averted with access to proprietary data. This is why all of ChatGPT, Claude, and Gemini push for agents and permissions to access your private data sources now. They will not need to train on your data directly. Just adapting the models to work better with real-world, proprietary data will yield a powerful advantage over time.

Also, the current training paradigm utilizes RL much more extensively than in previous years and can help models to specialize in chosen domains.

nopinsight··on GPT-5
Au contraire, AlphaGo made several “counterintuitive” moves that professional Go players thought were mistakes during the play, but turned out to be great strategic moves in hindsight.

The (in)ability to recognize a strange move’s brilliance might depend on the complexity of the game. The real world is much more complex than any board game.

https://en.m.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol

nopinsight··on Things that helped me get out of the AI 10x engineer imposter syndrome
How long would that be economically viable when a sufficient number of people can generate high-qualify code in 1/10th the time? (Obviously, it will always be possible as a hobby.)
nopinsight··on A.I. researchers are negotiating $250M pay packages
I agree. I just personally wouldn’t want to wander around exploring it continuously for months without more interesting work/goals. Even though cultures and geography may be wonderfully varied, their ranges are way smaller than what could be.
nopinsight··on A.I. researchers are negotiating $250M pay packages
What would you say if the IMO Gold Medal models from DeepMind and OpenAI turn out to be generalizable to other domains including those with hard-to-verify reward signals?

Hint: Researchers from both companies said publicly they employ generalized reasoning techniques in these IMO models.

nopinsight··on A.I. researchers are negotiating $250M pay packages
Sounds good in theory but I'd be bored to death in a month, at most two. Traveling the world...maybe good for a few more months and that's it.

Wouldn't you yearn for any more impact given how much that amount of resource could improve the lives of many, if used wisely?

nopinsight··on Hypercapitalism and the AI talent wars
impact != story points
nopinsight··on The ‘white-collar bloodbath’ is all part of the AI hype machine
My thesis is that this could lead to a booming market for “pink-collar” service jobs. A significant latent demand exists for more and better services in developed countries.

For instance, upper-middle-class and middle-class individuals in countries like India and Thailand often have access to better services in restaurants, hotels, and households compared to their counterparts in rich nations.

Elderly care and health services are two particularly important sectors where society could benefit from allocating a larger workforce.

Many others will have roles to play building, maintaining, and supervising robots. Despite rapid advances, they will not be as dexterous, reliable, and generally capable as adult humans for many years to come. (See: Moravec's paradox).

nopinsight··on FLUX.1 Kontext
Wondering if you could add “Flux 1.1 Pro Ultra” to the site? It’s supposed to be the best among the Flux family of models, and far better than Flux Dev (3rd among your current candidates) at prompt adherence.

Adding it would also provide a fair assessment for a leading open source model.

The site is a great idea and features very interesting prompts. :)

nopinsight··on Gemini Diffusion
...which is why top LLM providers' web apps like ChatGPT, Claude.ai, Gemini try to nudge you to connect with Google Drive, and where appropriate, GitHub Repos. They also allow the user/dev to provide feedback to revise the results.

All the training and interaction data will help make them formidable.

nopinsight··on Linguists find proof of sweeping language pattern once deemed a 'hoax'
Words are more than just symbols; they represent concepts and patterns we observe in the world, in our society, and inside ourselves.

Translation is only possible because we are all humans and have experienced broadly similar concepts, but there's a limit to it, especially in social milieu and in how we conceptualize ourselves in society.

To truly understand another people and culture at a deep level, you need to learn their native tongue and their living environment -- This is what I've internalized as a long-time learner and teacher of languages.

← PreviousPage 2 of 34Next →