Patrick Dinneen

I like data. I like art. I like running.

[Patrick Dinneen @patrick_dinneen]

@vboykis If this was introduced to just transparently give me better results I wouldn't mind. However, I really don't want to end up in an accidental LLM chat when searching for a specific account.

[Patrick Dinneen @patrick_dinneen]

@DJSTTDJ Do I want any length DJ Sabrina album? Always yes.

[Patrick Dinneen @patrick_dinneen]

@AnthonyBila Yes please for the photo book 🙏

[Patrick Dinneen @patrick_dinneen]

@SanlamInvest @SanlamInvest when will there be an update on the independent investigation? We are invested in the outcome and want to ensure that this isn't forgotten.

[Patrick Dinneen @patrick_dinneen]

@SanlamInvest It emerges that this isn't an isolated incident, as I'm sure you aware from tagged content on Instagram. Please consider the video footage in your investigation.

[Patrick Dinneen @patrick_dinneen]

@SanlamInvest @karlowitchy I appreciate that you're investigating the matter. As a matter of good faith I trust that the outcome of this investigation will be shared, otherwise this response rings hollow.

[Patrick Dinneen @patrick_dinneen]

@molly0xFFF Even more perplexing to me are the fanboys who will defend their hero to the bitter/twitter end. For what? I didn't even know this man existed until this deeply embarrassing fiasco of the last few weeks.

Gonna tap him for the keynote of Cringefest 2024

[Patrick Dinneen @patrick_dinneen]

@austinkleon Letterboxd has been a game-changer for me for this exact reason. It is useful to pair it with film-Tok and then follow those creators on Letterboxd. I've discovered so much more than anything algorithmic.

[Patrick Dinneen @patrick_dinneen]

@haxor @vboykis Another framing might be: when does using types add value?
If the audience is yourself then the answer to this lies in personal preferences. If the audience includes other engineers then it depends on applications of typing that could provide them with the most value.

[Patrick Dinneen @patrick_dinneen]

@haxor @vboykis Part of the value of using typing is as a communication aid to future readers and maintainers of the code base. If the types are so simple as to be implicitly obvious or the code is solely for you and you'll never revisit it then types might be unnecessary.

[Patrick Dinneen @patrick_dinneen]

@minkbazink They're easy if you know that you need to know anything about them (half the battle right there) and what you should do.

This is not the case for most other languages.

[Patrick Dinneen @patrick_dinneen]

@streamlit A Movie Recommender where you can find movies that are Similar to _Your Specified Movie_ but more like _X_ and less like _Y_

https://t.co/uMlNjFSKcX

[Patrick Dinneen @patrick_dinneen]

@alienelf In XP (now extremely out of fashion) they call it an iteration. https://t.co/R4uMGQ3lIB
Agile, as it is practiced now, seems a far cry from the original spirit of the manifesto.

[Patrick Dinneen @patrick_dinneen]

@Ben_Rausch @nannaventer All that money and still riding the bus!

[Patrick Dinneen @patrick_dinneen]

@GergelyOrosz Comments are invaluable for exactly this. Explaining non-obvious context (i.e. cannot be inferred from the code). Good comments explain "why" instead of "what"

[Patrick Dinneen @patrick_dinneen]

@franciscojarceo Rule #0: Understand your growth dynamics

A balance is required between the signals necessary to sustain operations (likely signals to VC via growth) and the outcome of focusing on delivering those signals (trade-offs wrt growth dynamics and implementation)

[Patrick Dinneen @patrick_dinneen]

@yishan Bring back Swatch Internet Time! https://t.co/kf1qYe8HRu

[Patrick Dinneen @patrick_dinneen]

@yishan In the extreme case (where training is poorly done) you may end up with a model where the sheer size allows it to represent its training data in a compressed form (overfitting) but may then generalise to non-training data poorly.

[Patrick Dinneen @patrick_dinneen]

@yishan The next big thing is consider the combination of:
1. Being able to make massive models (in terms of numbers of intermediate layers) and
2. Being able to train these massive models on staggering amounts of data.