HNHacker News
TopNewBestAskShowJobs

benjaminwootton

5,719 karma · joined October 10, 2010

linktr.ee/benjaminwootton
submissionscomments
benjaminwootton··on Ask HN: How do I train a custom LLM/ChatGPT on my own documents in Dec 2023?
The documents are encoded as vectors and stored in a database, so I suspect it would be effectively unlimited. You would just pay for storage and compute.

AWS OpenSearch has fairly good integration so you could look up costs for that. It’s not the cheapest AWS service to run and not exactly serverless as you pay by the hour.

benjaminwootton··on Ask HN: How do I train a custom LLM/ChatGPT on my own documents in Dec 2023?
To sing the praises of Bedrock again, it does have continuous pre-training as well as RAG “knowledge bases”. The former is based on JSON fragments and the RAG stuff is PDFs and other document formats.

With regards to its efficacy, I haven’t gone to production with it yet but I was reasonably impressed.

I uploaded 100 legal case documents to Bedrock via Claude and could push it pretty hard asking about the various cases and for situations across the knowledge base.

It did feel like it broke down and got confused at a certain point of complexity of questioning, but I still think it’s already useful as a “copilot” or search engine and surely it will only improve over time.

benjaminwootton··on Ask HN: How do I train a custom LLM/ChatGPT on my own documents in Dec 2023?
AWS Bedrock is fairly easy. You can do it in 5 or 6 clicks.

You have to upload your documents to S3, create a “Knowledge Base” then sync your documents into a vector database like OpenSearch or PineCone. You are then good to go via their playground or the AWS API.

I made a video here describing the process, check around 14 minutes in:

https://ensembleanalytics.io/blog/introducing-bedrock-knowle...

Bedrock is a decent product I think. All of the models in one place (apart from the big dogs from OpenAI) and a common API across them.

benjaminwootton··on “Yes” means “no”: The language of VCs
A less cynical take is that it allows you to run at a loss for a very long time as you build revenue and market share. This can be right up to the time that you exit.
benjaminwootton··on The new local data stack: Integrating Dagster, dbt and DuckDB
There are a bunch of collector tools for capturing application side events and sending them into your server.

The common stack is then to ingest them from Kafka into an OLAP datastore like ClickHouse or Pinot.

Then I think you are referring to what is sometimes called “user facing analytics” against that datastore.

We like to use a tool called Cube.dev as middleware between the app and OLAP database which gives APIs and a semantic layer.

benjaminwootton··on [dead]
Even without that. The timeline seems to be stock goes up, he sells his shares, then he leaves (which may or not be pre-planned). There is nothing suspicious in the timing at all.

It frustrating how the media can write things like this and potentially tarnish someone’s good name.

benjaminwootton··on Amazon-Unveils-Q
With the AWS LLMs, I noted the other day that they legally indemnify users against copyright claims. That was a big move I thought.
benjaminwootton··on GitHub: Can no longer search code without being logged in
I see lots of reasons being proposed, but my guess is that they want or need to avoid content being scraped by AI related companies and researchers.
benjaminwootton··on Ask HN: How to Deal with AI Anxiety?
Humans have moved up the stack since the wheel made it easier to move rocks around.

Even today with the globalised market, internet, mobile phones, ecommerce and all of the apps and social media to fill our time, we are still close to 100% employment.

We just seem to move onto higher level activities with each new generation of technology and find more stuff to do with our time and sell to each other.

Even if I’m wrong, knowing how slow big companies adopt disruptive technology, it would be decades before this is a real issue for developers.

benjaminwootton··on 7,000 pubs forced to close in last 10 years
I think Wetherspoons are unfairly maligned.

They give the market what they want in terms of cheap food and beer, and some of the venues are beautiful.

Most of them are fairly civilised during the week as they revolve around food and there are families present.

The city centre ones do get a bit rowdy on Friday and Saturday nights though.

benjaminwootton··on 7,000 pubs forced to close in last 10 years
It’s not a topic I’m particularly passionate about, but to take the opposing view:

Millions of people enjoy visiting pubs every week and the overwhelming majority don’t have an alcohol problem.

It’s an important venue for social connections at a time when there are less and less of those. They haven’t been replaced with other venues for many people, particularly the elderley who already struggle with social isolation.

They are also a big part of our culture and heritage at a time when that is also under threat from many angles.

The industry covers tens of thousands of jobs and captures boat loads of taxes both directly and indirectly.

And the annoying thing is that they are still viable small businesses, but the big pubcos and the taxes have squeezed the industry until it’s unviable to operate.

I’m not a big pub person, but I think it’s a shame to lose one of the few remaining pastimes which haven’t been replaced with an app.

benjaminwootton··on Developer account removed by Apple
I find that a little harsh. The prices are clear, and you can either pass or remember to cancel.
benjaminwootton··on GitHub Renamed Me (2014)
I do actually think this is reasonable. It's really annoying when the handle you want is taken by someone who hasn't used it for years.

Two years might be a little short though.

benjaminwootton··on Solving duplicate data with performant deduplication
This is the way the industry is going. Table formats such as Delta, Hudi, Iceberg stored on cloud object stores.

Though it works amazingly well, it is certainly slower than ingesting the data to be stored and manipulated in native formats.

benjaminwootton··on Solving duplicate data with performant deduplication
Indeed, you should still aggregate even on mergetree tables.

I'm not sure what is about the database world where people are happy to discuss their competitors and include either mistakes or misinformation. It doesn't seem to happen in other parts of the industry.

benjaminwootton··on OpenAI staff threaten to quit unless board resigns
I don’t know about OpenAI, but Ive been in a few similar business situations where everyone is in a good situation and greed leads to an almighty blowup. It’s really remarkable to see.
benjaminwootton··on OpenAI staff threaten to quit unless board resigns
There must be something going on which is not in the public domain.

What an utterly bizarre turn of events, and to have it all played out in public.

A $90 billion valuation at stake too!

benjaminwootton··on StyleTTS2 – open-source Eleven-Labs-quality Text To Speech
I’ve got a 64gb Mac M2. All of the openllm models seem to hang on startup or on API calls. I got them working through GCP colab. Not sure if it’s a configuration issue or if the hardware just isn’t up to it?
benjaminwootton··on Almost no one pays a 6% real-estate commission except Americans
How did you ever get to the situation where two people want to buy and sell one of life’s fundamental commodities, and 3 people get to insert themselves into the middle to the cost of thousands of dollars?

America confuses me sometimes!

benjaminwootton··on Almost no one pays a 6% real-estate commission except Americans
I would hate the high US property taxes. It’s like you never really own the property if you are being taxed on it continually. It must play havoc with retirement planning?
benjaminwootton··on Almost no one pays a 6% real-estate commission except Americans
In the UK, the market rate 1.5% paid by the seller on completion. Nothing to the buyer and rarely any fees for a failed transaction.

Then approximately £1000 in legal fees when buying or selling.

Then the big one which is stamp duty (a government tax) which is between 0 and 12% of the purchase price. Typically the lower end of that scale though. (https://www.gov.uk/stamp-duty-land-tax/residential-property-...)

benjaminwootton··on Tragedy of return to hostile offices
Is it though? Most companies prioritise their own self interest over nebulous “institutional investment” bogeymen.
benjaminwootton··on Ask HN: As a first-time solopreneur how would you go about hiring the first team
I think this is the top hiring mistake for startupS - hiring the big cheese from BigCo who turns out to be a total culture clash.

I do however disagree with your final paragraph. We had a few wantrapreneurs who were usually moonlighting and clearly didn’t want to be there.

benjaminwootton··on US Smartphone Shipments Decline 19% in Q3 2023 as More Americans Delay Upgrade
I just upgraded from a base model iPhone 13 (I think) to a 15 pro max. I cannot see a single difference apart from the strange pill like thing at the top of the screen. Even the battery life was fine on the old one. It really feels like an “emperors new clothes” upgrade.
benjaminwootton··on WeWork Goes Bankrupt
We were definetly spoilt with WeWork because the entire show was subsidised with SoftBank cash.

So many nice buildings in world class cities must have cost a fortune to establish, but the short term rental market cannot bare that much of a premium so it left us with a great product for only a slightly higher fee.

I agree the whole setup is nice, though the customer service leaves something to be desired.

benjaminwootton··on WeWork Goes Bankrupt
I had a baffling customer service experience with Spaces where I signed up, it took 2 months to get an access card that worked, then they claimed I’d signed up to something totally different. They would be one of the last companies on earth I’d ever sign up with again.
benjaminwootton··on Show HN: Ladder, open source alternative to 12ft.io and 1ft.io
It’s just a matter of opinion. I think Vercel were acting reasonably with all of the context we have. Things like terms of business are at the account level, not project level.
benjaminwootton··on Show HN: Ladder, open source alternative to 12ft.io and 1ft.io
I followed this drama on Twitter. The author was breaking the terms of service and creating DMCA support burden for Vercel. They had proactively been in touch with him a few times to reach a solution.

I think it’s quite reasonable that they blocked the account rather than the project. You wouldn’t have got that level of service from big tech.

benjaminwootton··on Consumers are paying more than ever for streaming TV each month
Streaming is a hot mess, but if you think about how many hours the typical person watches TV for, $50-$100 for an absolutely enormous on demand library with mobile access and HD quality etc isn’t that bad.
benjaminwootton··on GitHub 404 errors
Im relieved to see this here on HN. I nearly had a heart attack there.

Looks like issues with a bunch of services on their status page.

Friday afternoon deployment?

← PreviousPage 5 of 34Next →