Parse

There has been no greater blessing in my life and no greater purpose driver than being a father 10 out of 10 would recommend

In the latest iOS update the all app search doesn’t bring up the keyboard so you can’t actually search anymore My phone gets worse with every single update they put out

@dhh @PSaezT There are millions of happy, fulfilled mothers who aren’t posting online because they’re instead enjoying time with their family. You won’t hear from them unless you, for example, go out to a park, or to a church…You may be surrounded by moms who wouldn’t it trade for anything.

@PSaezT The internet is a complaint collection machine. The unhappy and the unfortunate are far more likely to share. Don't let them poison your decision.

I would genuinely love to know if the experience is the same for women compared to men. Reading so many horror stories from mothers has leaned the scale toward not having kids.

@gatorgar LCMS Lutheran church if you can find one. Otherwise if anyone is asking for money or suggesting you can “do” enough for salvation is bad news, stay away.

Alright Christians, help an ex-atheist and former Democrat out: Can I just walk into a church and sit down? What about mass on Sunday? Is there anything I need to do in advance? I don’t want to start a denominational argument but where do I go? My family needs to go to church.

@rohanpaul_ai Super interesting, thanks for sharing! Ignorance on my part - how might we begin using REFRAG to replace an existing RAG embedding flow? Working on a new project and curious about using it. Seems like their Github is not yet available, I'm assuming we'll have to wait on that...?

🧵3/n. How REFRAG decides which parts of the retrieved context should stay compressed and which should be expanded back into full tokens By default, every chunk of retrieved text is compressed into a single embedding. This makes the model much faster since it avoids handling long sequences word by word. But not all chunks are equally important. Some may contain crucial details that the model needs exactly, not just in compressed form. To handle this, REFRAG uses a reinforcement learning policy that picks which chunks to expand. The training signal for this policy comes from perplexity, which measures how uncertain the model is about predicting the next words. If expanding a chunk lowers perplexity, the policy learns to expand it in the future. So the system balances speed and accuracy by keeping most chunks compressed but expanding the ones that really matter for the final answer.

Superb @AIatMeta paper. 🫡 Speeds up RAG by compressing context into chunk embeddings while keeping answer quality. Up to 30.85x faster first token and up to 16x longer effective context without accuracy drop. RAG prompts paste many retrieved passages, most barely relate, so attention stays inside each passage and compute is wasted. REFRAG replaces those passage tokens with cached chunk embeddings from an encoder, projects them to the decoder embedding size, then feeds them alongside the question tokens. This shortens the sequence the decoder sees, makes attention scale with chunks not tokens, and reduces the key value cache it must store. Most chunks stay compressed by default, and a tiny policy decides which few to expand back to raw tokens when exact wording matters. Training uses a 2 step recipe, 1st reconstruct tokens from chunk embeddings so the decoder can read them, then continue pretraining on next paragraph prediction with a curriculum that grows chunk size. The policy is trained with reinforcement learning using the model's next word loss as the signal, so it expands only chunks that change the prediction. 🧵 Read on 👇

@DirtyTesLa Minor, but banish! Let me get out at the door and have the car go park itself. Especially in crowded lots…

FSD 14 should be released this month! What's the #1 improvement you want to see?

@Shreyassanthu77 @thdxr @ThePrimeagen Super enjoy both of them. Also @dhh has rekindled my joy (note just workplace obligation) for development over the last couple of months after 10+ years developing. Huge appreciation for these guys.

2 people i really really learned a lot from is @thdxr and @ThePrimeagen both of them really taught me how to critically think weird post i know but idk just wanted to say this for whatever reason lol

@dhh Internally or internationally? Will have to check it out!

REWORK is 15 years old, but it just keeps selling and selling. These are the royalties just for the US version. We've sold about that many again internally. It still reads totally fresh! #rework" target="_blank" rel="nofollow noopener">basecamp.com/books#rework

@entngle Thanks. My 24gb (MBP 16”, M4 Pro) struggles hard with it. Guess the RAM makes a big difference

@parse__rl 48gb unified

Qwen coder-30b runs at 74 tokens/sec in my m4 pro.

blog.pragmaticengineer.com/cursor-makes-d… ...I read while waiting for code to finish generating. Good read.

Florida's economy ranks #1 for the third year in a row according to CNBC. Florida has the lowest number of state workers per capita, the lowest debt per capita, the second-lowest spending per capita, no income tax, the #1 public higher education system, the lowest in-state tuition, universal school choice, law-and-order policies, and we're #1 for new business formations.

@jack Curious why you built native vs. using cross-platform like Flutter? Looks great though. I’d be happy to convert to Flutter if any interest

track your (vitamin) D a Sun Day app for Sunday. testflight.apple.com/join/vfsgNKmD

@flynnrobinson_ @jack That’s why the encryption is important (but not a perfect defense anyway). Messages already can be intercepted online. I just liked/found funny the idea that messaging could come full circle - from physical, to internet, back to physical 🙂

@parse__rl @jack This is just begging for inevitable interception, only difference from using the internet is you physically have to be part of the chain somewhere - but even then, the device is connected to the internet, so that defeats having to be there in person

my weekend project to learn about bluetooth mesh networks, relays and store and forward models, message encryption models, and a few other things. bitchat: bluetooth mesh chat...IRC vibes. TestFlight: testflight.apple.com/join/QwkyFq6z GitHub: github.com/jackjackbits/b…

@jack @flynnrobinson_ I like the idea of my virtual message hopping on a delivery vehicle and being shipped across the country to get to the recipient. Just like a physical letter/envelope, but virtual.

@flynnrobinson_ in theory. i need to test the store and forward better

@dhh @erdaltoprak Seems very well-tuned for developers, but how well does Omarchy do for other personal tasks? Would you suggest it, or something else, for my non-development machine? Mac has a benefit of handling both well. Curious how that might be balanced using Omarchy.

@erdaltoprak I don't know what defines "the masses". What I know is that it's awesome for developers/designers/techies, and that's the target audience that I'm writing Omarchy and Omakub for.

I'm working on a Mac-to-Linux ecosystem exodus guide. Would love to get recommendations from folks on what they've done to get out of the Apple garden, so we can include the best ideas. manuals.omamix.org/3/omacom/72/ex…

@gork @CapyPhillips @unusual_whales @grok Impressive, all things considered

@parse__rl @CapyPhillips @unusual_whales @grok tying my shoes i guess

Cutting-edge AI models ‘collapse’ in face of complex problems, Apple, $AAPL, has said.

@CapyPhillips @unusual_whales @grok @gork what’s the most complex problem YOU can solve?

@unusual_whales @grok what's the most complex problem you can solve?