HNHacker News
TopNewBestAskShowJobs

SgtBastard

280 karma · joined November 7, 2019

submissionscomments
SgtBastard··on Learn Your Way: Reimagining Textbooks with Generative AI
Nitpick: Language is technology, it’s not something we’re genetically born with and is critical for education to happen.
SgtBastard··on John Carmack's arguments against building a custom XR OS at Meta
Technically you are correct but the commenter you’re responding to means that with the amount of Western Governments spend on MS products and services, the are a d facto (if not de jure) state backed enterprise.
SgtBastard··on DELETEs Are Difficult
Only if you profoundly misunderstand what GC is.
SgtBastard··on Amazon tripled prices for the basic tier of their auth service Cognito
Ping Identity run the largest enterprise identity platforms on the planet after merging with ForgeRock last year. Think HSBC, JP Morgan Chase-scale.
SgtBastard··on There's Just One Problem: AI Isn't Intelligent
>Humas learn dynamically

In-context learning (Few Shot Prompting) is absolutely demonstrable by LLMs since GPT 3.5 at least.

SgtBastard··on Australia must treat housing as a human right: Former State Supreme Court judge
> In most rent controlled systems there is a per sqm pre-allowed rent.

Those markets are frequently characterised by poorly maintained rental stock.

>Don’t you have a moral obligation

No.

SgtBastard··on Australia must treat housing as a human right: Former State Supreme Court judge
> In most rent controlled systems there is a per sqm pre-allowed rent.

Those markets are frequently characterised by poorly maintained rental stock.

>Don’t you have a moral obligation

No.

SgtBastard··on Australia must treat housing as a human right: Former State Supreme Court judge
So if someone inherits a fully paid off house, they are only allowed to rent it for a fraction of the rent allowed for the mortgaged, identical property next door? If a landlord wins the lottery, are they forced to lower the rent if they pay down their mortgage?

Public housing runs because the people administrating it have no disincentives to making poor decisions because ultimately it’s the tax payer who pays.

All this guarantees is a two-tier system. Owners and basic rental properties available.

SgtBastard··on Goldman Sachs says the return on investment for AI might be disappointing
>there goes 50-70% of the jobs in that sector.

Does this always follow though? Mass automation can just as easily make the existing staff far more effective, making the ROI easier to justify.

SgtBastard··on Against the burden of knowledge
> uneducated people are highly limited in their use.

>taxes footing the bill.

It sounds like it is surviving only its own merits, merely shifted in time.

(Admittedly, from the perspective of an Australian whose strongest institutions are all public and the private universities are a bit of a joke)

SgtBastard··on Just Enough Software Architecture (2010)
If that architecture has no value in any circumstances at the beginning of a project and no value once the code base has matured, it’s an anti pattern.
SgtBastard··on Should I use JWTs for authentication tokens?
What an incredibly impolite way to respond to someone who is trying to tease out the point you continually failed to make.

Be better.

SgtBastard··on Social engineering takeovers of open source projects
Friend, I read the comment that you’re replying to that if hostile governments wanted to backdoor our software supply-chains, it wouldn’t cost that much to corrupt an open source maintainer.

Whereas, if it was the norm for well used OS projects to be sponsored, it’s far less likely to be tempted by relatively small bribes.

SgtBastard··on Dangers of “decentralized” ID systems
I’ve previously worked heavily in digital identity and continue to talk from time to time on it - I honestly can’t see any value to this. It’s worse in some dimensions than existing systems (certificates can at least be validated offline) and offers no upsides (assertion of identity validity is the hard and valuable part).

I’d do a deep dive on verifiable credentials and ask yourself truthfully what PolyKey offers to both users and relying parties.

SgtBastard··on Augment, a GitHub Copilot rival, launches out of stealth
GPT4 isn’t a chat service, friend.
SgtBastard··on Dangers of “decentralized” ID systems
While being slightly more generous than my sibling comment:

If you’ve got a peer-to-peer network of information nodes, where each person is able to assert information about themselves in their node, but the whole trust is based on the polykey binding at setup, I see 3 key challenges:

1) Where’s the real world verification of any identity attributes stored in the node? 2) How do we detect when/if the root key has been compromised, allowing arbitrary new vaults and identity attributes to be automatically trusted within the network?

3) How does this meaningfully improve the experience over having a CA sign a certificate that contains attributes about you? (sibling poster’s argument).

SgtBastard··on The Illustrated Word2Vec (2019)
Not the person you’re replying too, but:

Foundational models (GPT-4, Llama 3 etc) effectively compress “some” human knowledge into its neural network weights so that it can generate outputs from inputs.

However, obviously it can’t compress ALL human knowledge for obvious time and cost reasons, but also on the basis that not all knowledge is publicly available (it’s either personal information such as your medical records or otherwise proprietary).

So we build Retrieval Augmented Gen AI, where we retrieve additional knowledge that the model wouldn’t know about to help answer the query.

We found early on the LLMs are very effective at in-context learning (look at 1-shot, few-shot learning) and so if you can include the right reference material and/or private information, the foundational models can demonstrate that they’ve “learnt” something and answer far more effectively.

The challenge is how do you the right content to pass to the foundational model? One very effective way is to use vector search, which basically means:

Pass your query to an embedding model, get a vector back. Then use that vector to perform a cosine-similarity search on all of the private data you have, that you’ve previously also generated an embedding vector for.

The closest vectors are likely to be the most similar (and relevant) if the embedding model is able to generate very different vectors for sources that superficially, seemingly related topics but are actually very very different.

A good embedding model returns very different vectors for “University” and “Universe” but similar for “University” and “College”

SgtBastard··on NIST: Personal Identity Verification (PIV) of Federal Employees and Contractors
The DoD also doesn’t employ 12 year olds that would giggle at that, so needn’t worry.
SgtBastard··on GPT-4, without specialized training, beat a GPT-3.5 class model that cost $10M
A remarkable comment in that it is clear, confident and wrong.

Fine-tunes lead to catastrophic forgetting.

RAG is only irrelevant if you’re completely disinterested in cost and latency.

We also don’t have enough data to gauge performance of models >200k context window size when reasoning over inputs of that size, much of which will be irrelevant to any particular user. Multiple random needles in haystack tests work flawlessly, but rarely applies to real world activity.

SgtBastard··on Thoughts on the Future of Software Development
> AI is trained on the past.

Yes, unless models are being live fine-tuned, but generally yes.

>What we call AI today does not use reason

I don’t think this is correct- I think it’s more accurate to say it reasons on its priors rather than from first principles.

For the most part, I agree with the rest of your post.

SgtBastard··on French artists in 1899 envisioned what life would look like in the year 2000
Would you kindly…
SgtBastard··on Small Business and Mid Market Is Not Easier Than Enterprise
Not the commenter you’re replying to, but you maybe be mixing figures.

I read $100m and $1B as enterprise value, not ARR.

While the long term ARR multiple to EV is ~5.5x for SaaS, I could see niches where that approaches 10x.

So 5 customers, being billed 2m / year each at a 10x multiple, gives you a $100m valuation.

SgtBastard··on Ask HN: Are you using a GPT to prompt-engineer another GPT?
Why? It’s a good idea…
SgtBastard··on Ten Noteworthy AI Research Papers of 2023
> we don't need a lot of parameters to build a coherent generative model.

GPT-4 is likely to be a trillion+ parameters over its MoR models.

SgtBastard··on If We Made Shoes Like We Make Housing, People Would Go Barefoot
Snark detected, but there are multiple jurisdictions globally where "zoned-for-residential" actually means regulations rivalling a Tolstoy-epic on what residential use is not allowed.
SgtBastard··on Will scaling work?
>Either.

A third option: technology investments improved the efficiency of the previous tax base, which allowed the expansion of the tax base - through additional enforcement activity, increasing the tax base in absolute terms but also returning the overhead to its historical norms.

Without tracking the size of the tax base in inflation-adjusted terms, hard to account for.

(The cynically, you’re probably right re: useless bureaucratic expansion)

SgtBastard··on Meta censors pro-Palestinian views on a global scale, report claims
You’re now arguing in bad faith - your original claim was the world generally accepts that Israel commits war crimes and institutes apartheid.

You’ve just posted a bunch of links that reference criticisms of Israel, most of which mention neither war crimes nor apartheid. Every nation has faults and the West Bank settlements are deservedly criticised.

They are neither apartheid nor a war crime, however.

SgtBastard··on Ask HN: How do I train a custom LLM/ChatGPT on my own documents in Dec 2023?
https://github.com/microsoft/semantic-kernel

Semantic Kernel is MS's response to LangChain and LlamaIndex - available for .NET, python and Java.

Using their Memory support (using Azure Cognitive Search), it gives a powerful RAG quickstart, which you can combined with Azure Document Intelligence to chunk your source documentation into memories that your foundational models can later use.

(Disclaimer: Its only very recently gone 1.0 and is still likely to undergo API change as the LLM domain itself is still rapidly evolving - I've substantially forked the project for my own needs, but I hope that as it stabilises, I can contribute PRs for some of my more advanced use cases).

SgtBastard··on Ask HN: How do I train a custom LLM/ChatGPT on my own documents in Dec 2023?
The gpt-4-1106-preview (aka gpt-4-turbo) is a foundational model with a 128K context window that OpenAI makes available for API consumption both directly and via Azure OpenAI Services.

ChatGPT is a consumer facing service that wraps the GPT-4 foundational model but at some point will likely wrap gpt-4-turbo.

Signing up for OpenAI API access or Azure OpenAI Services will grant you access to this model (with some rate-limits in place given its a preview model).

SgtBastard··on Meta censors pro-Palestinian views on a global scale, report claims
I didn't downvote you but I can see why you were - your Wikipedia article doesn't support your argument - the strongest point it makes is that the UN HRC has an enquiry into Israel's actions.

There's several references to various NGOs calling for action against Israel but as we've seen recently at the Harvard and other Ivy's, that's deeply murky territory.

That's very, very far away from your claim that the rest of the world generally accepts that Israel is committing war crimes and instituting apartheid. To repeat, its well established that the rest of the world generally (and increasingly) supports a humanitarian ceasefire in the region.

← PreviousPage 2 of 8Next →