HNHacker News
TopNewBestAskShowJobs

carlossouza

488 karma · joined March 22, 2022

submissionscomments
carlossouza··on My solopreneur story
Exactly. This story is excellent... truly inspiring!

It reminds me of PG's mantra: make something people want.

carlossouza··on Outperforming larger language models with less training data and smaller models
Their researchers are great, irrespective of how successful their gazillion products might be.

Without their papers, there wouldn't be GPTs.

carlossouza··on Unlimited Kagi searches for $10 per month
I also stopped using Google, but I have replaced it with Perplexity.ai...

Have you ever used Perplexity? How does Kagi compare with it?

carlossouza··on The Tyranny of the Marginal User
Totally agree: it's hilarious because it's true.

Good startups target power users. By "good" I mean companies that find true PMF and grow exponentially at near-zero CAC (i.e. the ones that earn - not buy - their growth).

carlossouza··on The end of tweeted articles
> If your goal is visibility, you are well advised to tailor your messaging to the medium. That means skewing to the visual and away from the linked article.

Isn't that obvious?

carlossouza··on Show HN: Lantern – a PostgreSQL vector database for building AI applications
I'm using pgvector in production, mainly in a table with 500k-1M rows.

My main use case is to return search results with pagination: page 1 from 1-50, page 2 from 51-100, page 3 from 101-150, etc. (Think LIMIT and OFFSET).

After a lot of experimentation and help from pgvector's team, we discovered that, for this specific use case, IVFFLAT index is much faster than HNSW.

I looked at your documentation and only saw HNSW, no IVFFLAT.

What would be Lantern's performance for this specific use case?

Thx!

carlossouza··on TradingGPT: LLM multi-agent framework for enhanced financial trading performance
Renaissance is the benchmark for AI/ML applied to trading. Naturally, they are completely secret.

Euclidean, the fund that manages the personal wealth of some of Renaissance's key personnel, also uses AI/ML to trading... actually, to investing, more long-term.

They have publicly shared interesting research findings worth reading:

https://www.euclidean.com/data-posts-machine-learning

carlossouza··on Asking 60 LLMs a set of 20 questions
And what’s more data to a model if not patches that inflate performance?

The more data we use to train a model (or as you said, the more patches we use), the better it’s performance will be.

carlossouza··on Consensus: Use AI to find insights in research papers
Yes, probably. But isn’t GitHub Copilot, Jasper, Perplexity and other great apps as well?

Judging from the engagement in their slack channel, it looks like they are onto something.

carlossouza··on Market Map
As a digital art piece, it’s amazing!

(As a practical tool, not so much…)

carlossouza··on Is this a good book for me, now?
> Third, book recommendation systems based on your reading history can be helpful, but only so much.

Book recommendation systems should be a hybrid system with content-based filtering and collaborative filtering systems.

Reading history (lagging indicator) + browsing history (leading indicator) would form the basis of a content-based filtering.

The collaborative filtering part would ensure the system compares users with same reading patterns.

A system like this should address all author's concerns.

carlossouza··on Institutionalized Belief in the Greater Fool
I don’t think it’s that simple.

Kahneman and Tversky identified several cognitive shortcuts (heuristics) that people use to make judgments and decisions, which can lead to systematic errors (biases)... often leading to irrational decision-making. Their research explains several irrational decision-making processes that we demonstrate while investing (either in private or public markets.)

Thinking Fast and Slow is such a great book…

carlossouza··on Splitting the Web
Do you guys think the non-commercial web will ever achieve critical mass to tilt the scale in its favor and spark a chain reaction to drive flocks of people out of the commercial web?

A refreshing thing about the non-commercial web is the fact that it is not trying to monopolize my attention and not trying to hard-sell me anything.

More and more, I see people complaining about the big tech services... but that's a qualitative perception, certainly biased... I wonder if that's really happening.

carlossouza··on What we know about LLMs
Isn’t Copilot a killer app?!
carlossouza··on Show HN: I made a MailChimp alternative that connects to your database
Oh, one-time-fee products are so much better for users!
carlossouza··on Show HN: I made a MailChimp alternative that connects to your database
I'll update my beliefs after your data point :)

Congrats on EmailOctopus; it looks like a nice product!

carlossouza··on Show HN: I made a MailChimp alternative that connects to your database
I think there are some good ideas here, but the target audience imho is not clear.

Email marketing cheaper is a great attention grabber: kudos for the good copywriting.

Step 1: Link to AWS SES... humm, for that your target audience must be tech-savvy.

Step 2: Query your database... humm, linking my database to a strange system? No freaking way! But assume for an instant that this would be ok. For this, your target audience again must be tech-savvy.

Step 3: Create an email template using MJML... again, for tech-savvy people, not for the average digital marketeer...

Step 4: Review and send... ok, pretty basic..

English -> SQL. Here you got me confused. For steps 1, 2 and 3, your audience must be tech-savvy. This feature doesn't make any sense for tech-savvy people. It only makes sense for the average marketeer in a small team (or solopreneur) who doesn't know SQL. But this guy would never get through steps 1, 2 and 3!

See my point?

A tech-savvy person in a tight budget would develop his own solution. Imho, not your target.

An average digital marketeer or solopreneur who doesn't know how to code and is in a tight budget could be your audience. But for that, steps 1-3 must be no-code/for dummies.

The greatest thing about the idea is "email marketing cheaper".

carlossouza··on Why project-based learning fails (2018)
Clickbait article.

Implying that project-based learning fails is like saying the practice of learning to play whole pieces of music on piano fails (to use the same comparison the author has used.)

Staying on the author's comparison: to learn to play piano, one must practice exercises in isolation (deliberate practice, as Angela Duckworth calls it in "Grit," an amazing book) AND ALSO practice playing whole pieces of music.

The same is true for learning anything. Practice exercises in isolation PLUS project-based learning is the most effective way.

carlossouza··on Tell HN: WhatsApp Is Down?
Yes
carlossouza··on Never waste a midlife crisis
Wait until 40. It hits harder.
carlossouza··on Look ma, no React: I recoded my portfolio site with vanilla everything
> I became a dev when everyone was hiring for React, so React is what I taught myself immediately after bootcamp in the scramble for a job

A better education path (in a bootcamp, university, book, youtube, whatever) would start with the fundamentals and build it from the ground up... not the other way around.

(I just realized how lucky I am for being taught in the best possible order, from the ground up... first the basics... and how so many people didn't have this opportunity)

carlossouza··on Matrices and Graph
I disagree... there are more ways you can use this matrix to creatively extract information out of it.

For instance, you can normalize along the columns, and build a "recommender system" using matrix factorization.

With that, when a new person comes with a portfolio, the system will output a probability for this new person to acquire the other assets he doesn't have.

It's (the very basic) idea of how Netflix recommends movies.

carlossouza··on NotebookLM: An AI Notebook
Imagine you are a screenwriter, and you are writing a fiction script, reimagining events in the past (e.g. Inglorious Bastards)..

I wonder what this product would answer when asked how the WW2 ended.

How would this source-grounding mechanism work? Would it answer based on the truth/facts or on the source provided by the author?

carlossouza··on Calls to investigate tax sites for sharing financial information with Meta
“Don’t be evil” + “Move fast and break things” culture at its finest…

What the heck did these companies become!?

carlossouza··on UCLA professor refuses to cover for Dan Ariely in issue of data provenance
A) Our human brains are hardwired to trust others[1]. We are not programmed to question, probe, or doubt. We want to believe.

B) Moreover, our brains are also hardwired to storytelling[2][3]. When presented with a nice narrative, we feel rewarded.

C) Finally, to quote Charlie Munger, "Show me the incentive and I will show you the outcome". People in academia focus on the # of papers published.

Put (A), (B), and (C) together and we can explain news such as these.

[1] https://www.psychologytoday.com/us/blog/the-athletes-way/201... (hopefully this study is legit :))

[2] https://www.tadigital.com/insights/perspectives/art-storytel...

[3] https://cnlm.uci.edu/2018/12/04/story/

carlossouza··on Mark Zuckerberg: “We’re moving very slowly” (2012)
Curiously, Facebook announced buying Instagram a few weeks after this email.

And Instagram, under Facebook's leadership, apparently adopted Renren's "strong culture of cloning things" to the letter: first targeting Snapchat (2015-16), and now targeting Twitter...

Or as someone replied below: > "I personally am pretty in favor of this approach in life generally... I would love to be far more aggressive and nimble in copying competitors at the interface / last mile level"

carlossouza··on I am drowning in mutes: The current Threads experience
I can only imagine 2 use cases right now:

1. You are a celebrity or influencer and you are trying to “colonize” the place

2. You have a business that sells products or services to this people: celebs and influencers are your primary target

carlossouza··on Threads profile can only be deleted by deleting Instagram account, Meta says
Makes you wonder about Jack’s luck/skill ratio in Twitter launch.

After seeing Bluesky so far, it may be higher than what most thought…

carlossouza··on Ask HN: How to do market research for product?
Sure! One of the hottest ones is called Zak.

They just received a $15 million check from Tiger Global and other top VC funds in Latin America. Here are some news articles in English:

https://techcrunch.com/2021/11/10/digital-restaurant-startup...

https://www.bloomberglinea.com/english/tiger-global-leads-15...

https://www.crunchbase.com/funding_round/zakpay-series-a--b9...

Their website is in Portuguese only, but I think you can ask Google Chrome to translate it:

https://www.zak.app/

Hope it helps!

carlossouza··on Ask HN: How to do market research for product?
If the restaurant's idea is your idea, it is a good business.

But the real money is not on the monthly fee you will charge restaurants to update their menus. That you can give for free.

The money is in automating their purchases and delivery of ingredients they need to run their restaurants. If you already know what they serve and have a system running their backend, it's easy to do that.

And everyone in this business knows one of their main pain points is purchasing ingredients: it needs to be done frequently, fast, from a good supplier, preferably close to them, and at a good price. It's a hustle. They spend a lot of money, time and energy doing that.

There are fast-growing SaaS companies in Brazil making a lot of money in that area.

← PreviousPage 3 of 4Next →