gillesjacobs

Looks very cool!

I prototyped something like this with build123d for Python and Cursor + OCP VSCode plugin.

Build123d is too new with too little examples out there, unlike OpenSCAD. I can only get it to generate good code with largr reasoning models that access the latest docs. No fast iteration for build123d yet.

I need an LLM in my fax machine.

The team is currently skiing for two weeks so we'll have to get back to you on that.

The naming might be somewhat politically coloured but post training with quality data is the best case for uncensoring models: abliteration usually causes substantial drop in performance.

Too bad the created dataset is not open source, as that would allow to verify the objectivity of answers to make sure it is not just a different flavour of propaganda.

That dataset is strategically useful for Perplexity as many more CCP-censored Chinese models are sure to be released.

Using 03-mini-high + Search I get the right answer he was looking for:

  The species was first split at the subgeneric level by Gregory S. Paul in 1988—he proposed the name Brachiosaurus (Giraffatitan) brancai. Then in 1991 George Olshevsky raised the subgenus Giraffatitan to full generic status, so that B. brancai became Giraffatitan brancai. Later, a 2009 study by Michael P. Taylor provided detailed evidence supporting this separation.

I guess Mike Taylor will gracefully cede his point now?

It is very funny to me that someone would feel the need to complain about a niche factual error in pretrained LLMs without even enabling RAG. If you even know the basics about this field, you shouldn't be surprised.

Of course this was probably more about ego stroking his paleontological achievement than a thoughtful evaluation of the current state of LLMs.

Do you have any retrieval and generation metric scores (eg, KILT or NQ datasets)?

I know benchmark datasets are not the be-all-end-all, but a halfway decent score and inference-time, would really help sell your framework (or help engineers make the choice).

In any case, very cool work, I built a lot of RAG pipelines as freelance NLP engineer and I will try this out.

The technique of generating hypothetical answers (or documents) from the query was first described in the "HyDE (Hypothetical Document Expansion) paper". [1]

Interestingly, going both ways: generate hypothetical answers for the query, and also generate hypothetical questions for the text chunk at ingestion both increase RAG performance in my experience.

Though LLM-based query-processing is not always suitable for chat applications if inference time is a concer (like near-real time customer support RAG), so ingestion-time hypothetical answer generation is more apt there.

1. https://aclanthology.org/2023.acl-long.99/

Tis Leive, sis

My great grandparents had a friend die in this fire.

Then they and their kids had to go through it again 26 years later.

Safe to say that if you're using off-the-shelf character-based chunking, your AI app is not past PoC.

The parallel browsing has strong resemblance to Xanadu from Ted Nelson's, which was a proposed web protocol and browser.

But I am vague about Xanadu's browser as I only seen it as a brief flash in Werner Herzog's Lo and Behold, Reveries of the Connected World.

You would think policy makers have an obligation to track this, but they don't. There are no studies, KPIs or attempts at tracking effectiveness of subsidy policy.

The burden of proof should be on the policy maker and subsidizer to ensure the purported goals of spending tax many are met.

My wife works in public procurement for research, but there too suggesting such a thing is taboo.

This is of course by design: subsidies are a political tool for passivation of opposition, soft bribery and economic channeling of economic and societal outcomes. The less oversight and transparency the better.

The Flanders region only got a publicly and centralised database of (some) subsidy channels in 2019, after decades of liberal and right-leaning people lobbying for transparency.

There is no such initiative for our federal government.

A free market of voluntary transactions solves this problem

According to that calculator my CA tax burden ratio would be 29.7% so seems the websites I read were not far off the mark.

Pray tell, what enlightened incorruptible entity would enforce and organise UBI?

I am a self-employed freelancer and I have done all fiscal optimization possible in BE and my effective tax burden is 37%.

From what I find in Google about CA's total optimized tax burden for a one-man freelance company it is about 28%. And a lot more buying power too probably, but that is hard to compare in.

Of course this is personal income from labour via one-man companies ("net take-home"). This does not take into account VAT, capital gains taxes, etc.

Additionally, for laws like this that specifically take away democratic control this argument is entirely moot.

Any claimed benefit or justification is entirely nullified for regular people if their oversight and transparency processes are being negated.

You hit upon the "locality principle of representative democracy": the more non-local elected representatives are in their excertion of power, the less oversight by the electorate. EU Law is seperated from the electorate by multiple layers: electorate > national parties > national parliament > national government > Council of the EU > European Parliament > European Commission > EU law

Each layer dilutes democratic input and accountability. By the time you reach EU law, the average voter's influence is homeopathic at best.

Localists would argue since a federation is highly non-local it shouldn't even be attempted because it leads inherently to non-representative centralisation of power.

This multi-layered separation explains why EU citizens feel disconnected from Brussels decision-making. It's a game of democratic telephone where the message gets garbled with each step and the centralist power machine inserts its own interests.

You're taking this very personally and in the public domain even though it will hurt your company.

Maybe there gets a point you become so wealthy, that ego is all that matters. But you're supposedly a smart guy who cares about his products, community and company, why not take the practical and economically safe route of duking it out in court?

> But calling the EU not being democratic is going too far

Unelected commissioners draft laws, while elected MEPs rubber-stamp with barely any changes. The Council's backroom deals make smoke-filled rooms look transparent. Try explaining the Ordinary Legislative Procedure to the average voter - you can't.

And let's not forget the ritual of ignoring referendums that don't align with the "ever closer union" mantra. When lobbyists have more sway than citizens, you've got a problem. The EU is a masterclass in obfuscating accountability behind layers of bureaucracy. It's democracy laundering, plain and simple.

1. Three years ago, you liked some tweets of someone who you didn't know was a political seperatist. Now, his group had been classified a terror organisation.

2. You went to buy some acetone as a solvent for a project.

There is now material evidence and intent for a terrorist attack using TATP bombs.

You and your family are on a no-fly list and are being watched.

All digital platforms have to automatically turn over all your personal data to intelligence services. They will share it with every other intelligence service in the union.

During automated analysis, a CASM detector finds bath-time pictures of your child. A criminal trial for the production of child pornography is initiated.

Like many I am looking for an exit too, at least fiscally. Right now I am -like many here- territorially bound by family and real-estate.

But the negative economic death spiral has set in here. More people work for the public sector directly and indirectly than the private sector, and the tax burden is ever increasing while services received are dwindling.

Ah yes, I am truly misanthropic for questioning my 60% tax burden mainly to the benefit of parasitic bureaucrats.

In Belgium art grants are basically determined by committees of peers organized by largely politically independent bureaucrats at the many different levels of government we have.

It's all navel gazing, entrenched interests and nepotism. Effectively the art sector is given tax money to perpetuate their own interests with limited democratic control.

Coincidentally, the "corporate" media in Belgium is also heavily subsidized (as in a significant amount of their revenue is subsidies).

And we also have public media. Yet no credible evidence of the effectiveness of endless subsidies.

If we're going to namedrop economic laws like it means something: Parkinson's law: the art and media sector have become part of the self-perpetuating ineffective bureaucracy.

In Belgium, it is relatively easy to get "artist-statute" which is a subsidized higher monthly welfare income (Basic Income for artists). Belgium runs on the "subsidize-and-conquer" paradigm, the tax burden is one of the highest in the world. The government keeps lower class happy with welfare, the upperclass with culture and mainly business subsidies. The productive middleclass is effectively squeezed.

The art sector here is mostly publicly funded. This too has advantages for our gov: no artists will criticize the subsidizer of their lifestyle, so no real anti-authoritarian culture takes hold. Don't bite the hand that feeds.

You'd think Belgium would have some impressive world-class artists, since so many can work unbothered by the oppressive forces of market trends. Not surprisingly, mediocrity, irrelevance and low-output are the norm.

> Did you put any copyright notices in your work? If not, it may be too late, because they can continue to distribute the old version that did not have copyright notices.

This is not how copyright works. Without a license, default copyright law applies, and no one can make any copy of the code and profit from it, even if published on Github. Copyright protection is automatic under the Berne Convention, implemented by the US Copyright Act and EU Directive 2001/29/EC, meaning no registration or notice is required for protection.

This is also why chip implants from eg Dangerous Things with MIFARE were desirable: you could clone old MIFARE chips this way using some tools.

Sadly, neither my gym or work access card were cloneable even though they are MIFARE Classic. So I did not end up getting an implant.

Pretty good, it found some useful references I missed in Google Scholar and Arxiv. Looks promising, will use it more.

The algorithm I use to get a final ranking from multiple rankings is called "reciprocal ranked fusion". I use the implementation described here: https://docs.llamaindex.ai/en/stable/examples/low_level/fusi...

Which is the implementation from the original paper.