teleforce

Am I missing something, if you a paid subscriber, LLM service providers OpenAI/Google/etc by default will not trained on your input data unless explicitly allowed by the users.

If this is not enabled by default then it need to be the default policy within EU by the enforcement of GPDR, for example.

>Essentially, we’ve had no major forms of innovation in memory for three decades – SRAM, DRAM, and Flash as the only memory types in broad deployment

Not sure about that, does 3D XPoint or Optane memory count as new type of memory innovation, it's a joint venture between Intel and Micron [1].

[1] 3D XPoint:

https://en.wikipedia.org/wiki/3D_XPoint

>We would need all the intellect in the world to get the interface narrow enough to be usable,

It seems Dijkstra is describing potential calability of LLM model trained on published collective human knowledge enabled by the Internet

>and, in view of the history of mankind, it may not be overly pessimistic to guess that to do the job well enough would require again a few thousand years

Since the article was written in 1978 (about 50 years ago) and from the time the computing resources of CPU and RAM has increased exponentially until the LLM first appeared in 2020s, it seems the thousand of years prediction is merely several decades accelerated by the rapid advances of computer and communication systems technology spearheaded by companies like Nvidia.

Very apt comparison and analogy.

Fun facts, Abrahamic religions that are based on the Bible and Quran both claiming human are made from clay [1],[2].

In Hebrew name Adam, the very first human, comes from the word adamah, which means "ground," "soil," or "red clay".

In Arabic Adam, comes from the word adeem, means the surface of the earth or dust, reflecting the Islamic and Abrahamic tradition that Adam was created from the clay or dust of the ground.

By extension in Hebrew and Arabic the word Adam also means mankind and humanity.

[1] Genesis 2:7:

https://www.biblegateway.com/passage/?search=Genesis%202%3A7...

[2] Quran 7:12:

https://quran.com/al-araf/12

Why so negative?

History has shown that a very potent tool can improved and changed industry entirely for the better. People now seems to be more negative than when atomic bomb was invented.

Looked what happened when crane was invented and not only more houses are built faster but also high rise building suddenly become feasible with more human being employed in the construction industry than ever.

Please check this interview with Lewis Nicolle who created his own game engine with Dlang for his new game Art of Reflection, and why Dlang very much suitable for game engine development [1],[2].

[1] Why D Programming Language?

https://www.youtube.com/watch?v=teWQbYvPBTg

[2] Art of Reflection:

https://store.steampowered.com/app/2290770/The_Art_of_Reflec...

>Coconuts grew in their native Maldives, in India, and perhaps parts of western Africa and the Middle East

None of the countries mentioned have coconut as native, its originated from the Malay Archipelago (East Indies) area. The common name for coconut from Philippines to Hawaii across the Pacific islands is nyiur/niu, and most probably brought along to Polynesian islands by the early settlements who speaks Malayo-Polynesian or Austronesian languages.

Fun facts, coconut name is a misnomer and it's originated from the Portugese, one of the first European in contact with coconut who thought it's a form of nut but actually it belongs to drupe or stone fruit family [1]. Specifically coconut is from the palm family and its fruit is mainly the kernel. Oil palm has similar kernel fruit but it's native to Africa.

[1] Nightmare Fruit: How the Coconut Got Its Name.

https://medium.com/@sixtybolts/nightmare-fruit-how-the-cocon...

I'm working in the field for CVDs early detection of arrhythmia and ischemia with just ML. We managed to get the elusive inter-class 100% in recall/sensitivity for both which is excellent for expert cardiologist-in-the loop by using the public dataset given the the current EU/US standards of SCORE-2/PREVENT screening for CVD with very poor around 50% recall/sensitivity [1]. The significant improvement mainly in ischemia detection due to the notorious silent ischemia that's not apparent even after CT angiography or by cardiac electrophysiologist (cardiologist) inspection.

For CVDs if you can detect and confirm early it can "tremendously improved outcomes for patients" due to the current avauability of new effective drugs and treatments.

Hoping that we can get our techniques tested on real-time data since it's super difficult to find relevant physicians/doctors collaborating on the new screening techniques on anything although for our case it's non-invasive. Perhaps need to pivot for early detection of CVDs in diabetes patients instead in order to get our hands on real-time ECG data.

[1] Risk stratification for cardiovascular disease: a comparative analysis of cluster analysis and traditional prediction models:

https://academic.oup.com/eurjpc/advance-article/doi/10.1093/...

>What you need is for your LLM to be able to understand enough context to be able to make a change with as few token as possible, so if your code isn't expressive enough

Recently there's an article on language plasticity in the era of AI/LLM and why D language is very well suited for this era [1].

Perhaps we need a proper benchmark similar to Beaver but for AI assisted coding for different programming languages instead of Text-to-SQL [2].

[1] Language Plasticity is More Important Than Ever:

https://blog.dlang.org/2026/08/10/language-plasticity-is-mor...

[2] BEAVER: An Enterprise Benchmark for Text-to-SQL:

https://beaverbench.github.io/#overview

>Since its introduction to the Linux kernel, netkit has been deployed in production at scale by Meta [15] and ByteDance [12] to improve container networking performance.

In "How we trained Clef" section Cloudflare mentioned that they initially based their Jev-like system on DiffussionGemma (DJev) model but then changed to Qwen with post training for Clef, but never mentioned any reason and justification for the change.

Although they included the benchmark against DJev and Clef is better, perhaps if you can test it to see the real-world performance.

Care to mention examples of deterministic options available?

This is definitely one of the best blog articles that I've discovered from HN on AI/LLM.

I share most of the article opinions and sentiments but there's one thing that escaped me regarding the contention of "English is the new programming language", and this probably his main beef with DHH.

English is not even a decent language in general nor it's a good language for communicating let alone for coding. For communicating the best language is Arabic, period.

But for coding and knowledge representation in general there's a deterministic English namely controlled English [1],[2],[3].

Perhaps someone need to make the proper language abstraction for LLM, English -> Controlled English -> Prompt -> LLM -> Controlled English, thus it will be much more consistent and deterministic rather than just English -> Prompt -> LLM -> English.

[1] Integrative Use of Information Extraction, Semantic Matchmaking and Adaptive Coupling Techniques in Support of Distributed Information Processing and Decision-Making (2011) [PDF]:

https://eprints.soton.ac.uk/272677/1/p4-acita.v5.pdf

[2] Controlled English to facilitate human/machine analytical processing (paywalled) (2013):

https://www.spiedigitallibrary.org/conference-proceedings-of...

[3] Controlled English for Knowledge Representation (2010) [PDF]:

https://attempto.ifi.uzh.ch/site/pubs/papers/doctoral_thesis...

>this is entirely purposeful

Please check Norman Finkelstein's latest book namely Gaza's Gravediggers on this matter [1].

He also has a very telling and chilling interview recently on Palestine and Gaza [2].

[1] Gaza's Gravediggers: An Inquiry into Corruption in High Places [book].

[2] Norman Finkelstein Reveals Shocking Discoveries While Researching History of Israel’s Gaza Obsession [video]:

https://youtu.be/Jmv2h1WYnNo

The OP article didn't mentioned it, but there's LocalJev based on DiffusionGemma depends on unmerged vLLM [1].

[1] A local, Jev-compatible POST /v1/systemone API written in TypeScript for Bun, backed by DiffusionGemma through an OpenAI-compatible Chat Completions endpoint.

https://github.com/githubnext/localjev

>acting on their own

Just like the Terminator movie /s

Once you understand how to write a program get someone else to write it - Perlis.

[1] Epigrams in Programming:

https://engineering.yale.edu/academic-study/departments/comp...

Thanks for the book recommendation.

Didn't know that MIT press features open access books.

In response to GP post, it's a valid concerns to programmers (to some extend to machine in form of AI/LLM code generation cost see ref [1]), and certainly we have nice things in programming language as D language aptly shows.

>Aspects of language design that are way more important are: the semantics, including type system, memory model, concurrency model, module system and so on; tooling; performance; FFI and OS interface; and lots of other things.

D have pretty much all of aspects of your main concerns and being much better than Rust and Go [1].

[1] Language Plasticity is More Important Than Ever:

https://blog.dlang.org/2026/08/10/language-plasticity-is-mor...

>Go grandmaster Shin defeats AI KataGo with a two-stone handicap

English is not my first language but for clarity perhaps the title should be:

Go grandmaster Shin with a two-stone handicap defeats AI KataGo

>All our moral codes and social norms we live by are purely human invention, derived from things that happen to work to produce a somewhat functioning society.

Religious people would disagree, and the evidences are all to see, hear and ponder from their sacred holy books. But the main questions are that which holy books are truly sacred, meaning that purely and utterly God words. Otherwise it's human innovation/exaggeration or something in between (i.e corrupted God words).

Based on these holy books, religious people adhere and follow their prophets teachings and actions to the best of their ability since the prophets are their fellow human beings and not angels send from the heavens.

Earlier this year France has pull out their last remaining gold held in US amounting to 129 tonnes moved to Paris [1].

[1] France pulls last gold held in US (361 comments):

https://news.ycombinator.com/item?id=47658146

Just wondering on the reported accuracy of the SQL queries generated by SQL based on RAG. Reports have been not encouraging and also confirmed by these results but perhaps your methodology is different from these reports [1],[2],[3].

>On this benchmark, a pure LLM generated an accuracy score of zero. Adding RAG, prompt engineering, and agentic AI raised accuracy to the 10+% range.

[1] Any text-to-SQL benchmark should address difficulties of real-world data stores (acm.org) (21 comments):

https://news.ycombinator.com/item?id=49013995

[2] If You Think You Can Do Real-World Text-to-SQL:

https://cacm.acm.org/blogcacm/if-you-think-you-can-do-real-w...

[3] BEAVER: An Enterprise Benchmark for Text-to-SQL:

https://beaverbench.github.io/

Perhaps you can try DECT NR+ the new reincarnation of DECT and also backward compatible with the original DECT [1]. It can also do roaming.

It's under 5G standard but for non-cellular connectivity that means you can use it without base-station like Wi-Fi direct. It's geared toward IoT but its data rates are from 1 Mbps up to 1.3 Gbps depending on the operating frequency and the type of modulation being used [2].

[1] DECT NR+: A technical dive into non-cellular 5G (30 comnents):

https://news.ycombinator.com/item?id=39905644

[2] Technology: DECT NR+ [pdf]:

https://www.vdma.eu/documents/d/group-34568/technology_dect-...

CT scans cannot detect some of CVDs since it mainly cover the vascular vessels part of the CVD not the function of the cardio engine (heart). The test procedure is only recommended if someone has CVD symptoms but not for screening. In most of the countries it will take months to perform the procedures and get the results due the scarcity of the facilities and the severe shortage of the cardiologist to perform the diagnostic with ratio to population of around 1:100,000.

ECG is the only one technique that cover comprehensive CVDs screening and it's purely non-invasive (no injection necessary). If someone has blocked arteries or even genetically malform heart like Hypertrophic Cardiomyopathy (HCM) it will show in the ECG traces, that can be cross checked with other relevant CVD diagnostic techniques. The main problem is that the screening GP and even the cardiologist has trouble performing the correct interpretation. This is where advanced signal processing and analytics, ML, AI, etc can really help, but somehow the heart fraternity like AHA is very much still in denial on this matter.

Please check the short overview article from Prof. Friedman on ECG based CVD detection in my other comment, ironically commemorating on the 100 years of AHA.

>echo cardiogram

The main advantage of ECG compared to echo is that it's represent the functionality of the heart and overall CVD rather than structure like echo.

CT scan is not for screening and risk prediction, it's normally recommended after discovery of CVD related symptoms.

Having said that even CT cannot detect some of CVDs since it mainly cover the vascular part of the CVD not cardio engine (heart).

ECG is the only one technique that cover comprehensive CVDs screening and it's non-invasive.

That's the main contention point, believe it or not current standard for heart diseases risk prediction method do not include the 'elephant in the room' ECG.

There are many related research work with excellent results, patent, etc by biomedical researchers around the world that I'd care to mention, including yours truly. Biomedical researchers even has yearly global competition organized so called CINC that used ECG as one of the main input for heart disease risk prediction since it's readily available [1].

In particular check out work by Prof. Friedman on ECG [2]. Somehow his excellent work is mostly being ignored by AHA and their counterparts because apparently these standard risk prediction models in US and Europe are not including it as part of their parameters.

[1] Computing in cardiology (CINC):

https://cinc.org/

[2] The Electrocardiogram at 100 Years: History and Future:

https://www.ahajournals.org/doi/pdf/10.1161/CIRCULATIONAHA.1...

>For non-invasive heart disease risk prediction nothing beat ECG, period.

Emphasis on non-invasive diagnostic screening as invasive testing like angiogram need to be operated by specialist and can take months to be arranged, and only done after incidents e.g heart attack.

ECG is excellent for generic top level CVD anomaly conditions for examples arrhythmia and ischemia.

For non-invasive heart disease risk prediction nothing beat ECG, period.

Somehow American Heart Association and its European counterpart are in denial, and still pushing dinasour screening mechanism with very low accuracy for heart disease risk prediction.

The standard risk model for CVD based on PREVENT (US) and SCORE-2 (Europe) like parameters are very poor as reported in the recently published paper on the their accuracy performance by the Swedish team [1]. As all CVD risk stratification with cardiologist review (expert-in-the-loop), the most important accuracy metric is sensivity/recall (avoiding false negative that will escape review) of PREVENT and SCORE-2, 26% and 48%, respectively.

The paper alternative proposal increased the sensitivity to 58% by performing clustering instead of conventional regression models as practiced in the PREVENT and SCORE-2.

These type of models including the latest proposal performed very poorly as indicated by their otherwise excellent and intuitive display of graphical abstract results [1].

[1] Risk stratification for cardiovascular disease: a comparative analysis of cluster analysis and traditional prediction models:

https://academic.oup.com/eurjpc/advance-article/doi/10.1093/...

Check this DLLM D project, a minimal, clean coding agent built directly on llama.cpp without Python, bindings or overhead [1].

The blog on the DLLM development [2].

[1] DLLM:

https://github.com/DannyArends/DLLM

[2] Teaching an AI to Know Itself: Building a Local LLM Agent in D:

https://blog.dlang.org/2026/06/07/teaching-an-ai-to-know-its...