wenc

9,688 karmajoined 10 tahun yang lalu

comments

8 hari yang laludiscuss

I did a search and found that the satellite internet market size is $16B today and $38B in 2031 (5 years from now)

https://www.mordorintelligence.com/industry-reports/satellit...

But I also think that focusing on residential internet as the only market might be thinking too small.

There's aviation, maritime, defense, telecom/enterprise backhaul, remote industrial (oil rigs, mines, etc.), and those guys are are not paying $135/month.

This might unlock new applications, like remote sensors and autonomous devices that are out of coverage areas today.

John Deere's farming equipment for instance is already on Starlink, and those things are basically computers on wheels.

The only issue is that satellite internet needs line of sight to the sky. Underground/undersea applicatons are basically out.

22 hari yang laludiscuss

As someone from Chicago (actual Chicago, south side, not the suburbs), randomly talking to strangers is what we do.

We're talking to strangers at the bus stop, at the grocery check out, or just wherever. It's just phatic conversation, nothing needs to come of it. Chicagoans aren't just friendly, they actually love the art of the conversation -- every conversation is a chance to put in the reps.

But the minute you step into the suburbs, this habit disappears.

28 hari yang laludiscuss

Just to add to your Texas comment: there are a few larger states that have distinct hubs where people think so differently, that they might as well be living in different states. People who haven't lived in these states don't really have a good feel for how different the internal cultures can be.

California: we know how SoCal is culturally worlds apart from Norcal, and both are worlds apart from inland California.

North Carolina: culturally Charlotte ≠ Research Triangle ≠ Greensboro-Winston Salem-High Point. There is no single NC culture.

Florida: the stereotypes exist, but I've visited different metros in Florida and they couldn't be more different. South Florida (Miami) is very Latin while the panhandle (Tallahassee, Pensacola) couldn't be less Latin -- it's mostly southern culture. Orlando, Tampa are also way different.

Florida in fact shouldn't be governable -- every part of the state has different interests. Yet it somehow works.

bulan laludiscuss

Attached is the writing.md I use to steer Opus 4.8. Prompt:

  "use the writing.md steering on x.md and loop until all LLM traces are removed".

I ran it on TFA and Pangram flagged it as LLM generated but Claude Fable couldn't definitively tell.

-- writing.md ---

# Writing Rules (MANDATORY)

## Banned Words and Phrases

Never use: "incredibly", "extremely", "absolutely", "fundamentally", "dramatically", "crucial", "vital", "powerful", "robust", "elegant", "seamless", "cutting-edge", "game-changing", "groundbreaking", "It's worth noting", "Importantly,", "Interestingly,", "Let's dive in", "At its core,", "At the end of the day,".

No exclamation marks in technical writing. No contractions in formal writing.

## Banned Sentence Structures

1. *Semicolons joining independent clauses.* Do not write "X does A; Y does B." Use a comma + conjunction that names the relationship: ", while" (contrast), ", and" (addition), ", so" (consequence). Semicolons hide the logical link and sound artificially balanced. 2. *Label-colon-explanation.* Do not write "The key insight: ..." or "The limitation: ...". State the point directly or use "is that" phrasing. 3. *Colon after a bolded term.* Do not write "a *rollout engine*: a lightweight...". Use a comma. 4. *Sentence fragments as assertions.* Every claim needs a subject and verb. "No gap at any ρ." → "There is no gap at any ρ." 5. *Em-dashes joining independent clauses.* Do not write "X does A — Y does B." Use a comma + conjunction. Parenthetical em-dashes ("the policy — trained offline — cannot adapt") are fine. 6. *Tricolon lists of near-synonyms.* "It does not X, Y, or Z" is padding unless each item is genuinely distinct.

## Banned Rhythms

1. *Staccato sequences.* Two or more consecutive short declarative sentences of similar length. Join them with a conjunction or subordinate one. A single short sentence standing alone for emphasis is fine and often good. Do not eliminate it. 2. *Formulaic layout.* Do not produce: intro paragraph → three bullets → summary paragraph. 3. *Gratuitous parallelism.* Do not force list items into identical grammatical form if it makes them sound robotic. 4. *Saying it twice.* If you stated a fact, do not rephrase it from another angle in the same paragraph. One clear statement is enough. 5. *The negation-correction reversal.* This is the move where you deny one candidate and assert the real one. Surface forms to match: "not X, but Y"; "it isn't X, it's Y"; "X was never the point, Y was"; "for me X, for them Y"; the comma-tag "Y, not X" ("sanctioned, not stolen"); the "not so much X as Y" form; and the gapped version where a stranded verb delivers the pivot ("Hours aren't the bottleneck. Attention is."). One reversal at a genuine turning point is good writing. The structure is not the problem. The density is.

   Detection is a whole-document pass, not a per-paragraph glance. Read the entire piece and mark every sentence or sentence pair that negates one thing to elevate another, including the comma-tag and gapped variants above. Count the marks. More than one per ~300 words, or more than three in a short piece, means the reversal has become the default sentence engine, which is the machine tell. A single dense paragraph with two stacked reversals also counts.

   Fix by thinning, not by deleting all of them. Keep the two or three that land on the strongest turns. Rewrite the rest as plain declaratives that state the point with no foil ("The bottleneck is attention now." instead of "Hours aren't the bottleneck. Attention is."). Removing every instance flattens the voice, so the aim is to make the survivors rare enough to regain their force.

6. *Repeated hedge adverbs.* A softener like "almost", "somewhat", "rather", "a bit", or "fairly" used more than once in close range becomes a tic. Keep at most one, and only where it earns its place.

## Positive Rules

- Active voice. Use "we" and "our". - Concrete nouns and verbs. "The model overfits after 50 epochs" not "exhibits suboptimal generalization characteristics." - Plain English. Use technical terms only when they carry meaning plain English cannot. - State consequences, not meta-commentary. "The policy has no lookahead" not "training compresses multi-period consequences into a single-step mapping." - One sentence that advances to the next thought beats two sentences restating the current thought. - State assumptions when uncertain. Do not hedge-stack ("it might be the case that perhaps..."). - Not every paragraph needs a topic sentence or a concluding sentence. - Do not resolve the ending with a tidy bow. A piece may close on an open question, an admission, or an unresolved tension. Summary endings that restate the thesis read as machine-generated. - Do not over-smooth. Removing every short sentence, every parallel, and every fragment flattens prose into uniform medium-length sentences, which is itself an LLM smell. Fixing a tell should not cost the voice. - When editing existing text, match the density and register of surrounding paragraphs. - Direct and conversational register, but no contractions in formal writing. Personal essays and conversational pieces keep their contractions; the no-contraction rule applies to formal and technical writing only.

bulan laludiscuss

I have a steering .md file that instructs Opus how not to sound like an LLM when writing prose (I write prose in my IDE with Opus). The steering is specific to me, but I've found that giving Opus rules like eschewing punchy journalistic sentences ("Not because X. But because Y. And that matters."), varying sentence lengths and avoiding staccato sounding clauses go a long way in smoothing out LLM smells in writing (at least according to me).

Aside: different LLMs sound different too! ChatGPT is the worst offender for LLM-sounding writing and needs the most smoothing, but Claude (web) actually sounds like a humanities major from the get-go.

2 bulan yang laludiscuss

I can attest to this. I work in Midtown Manhattan. You'd think walking around meant getting distracted by the all the activity around you that you'd forget about the problem you're trying to solve.

But I've found that distraction is the catalyst. Creativity for me comes when I focus on something else for a while, not grinding on the same problem with unwavering focus.

2 bulan yang laludiscuss

People often say Mandarin and Cantonese are like Spanish and Portuguese, but that undersells how different they are.

Your example of Spanish and French is more accurate -- same language family, but different grammar and vocabulary.

I offer German and Dutch as another example pair -- same language family as well, but different enough that no one will say "oh they're just different dialects". Dutch is an example of what happens when a Germanic language (Low Franconian) gets it's own state.

2 bulan yang laludiscuss

Category theory is rarely useful by itself, but it can be a mental scaffold when designing things like query languages. Microsoft's LINQ dsl within C# used category theory ideas to ensure consistency. That said, the applicability surface area in practice is typically quite limited in my experience. It's like formal methods -- elegant in practice, but a good problem fit is often rare. It's like writing a LEAN proof for your web app -- rarely needed, but if your web app needs a high degree of correctness, then indispensable.

This is John D Cook's take:

Category theory can be very useful, but you don’t use it the same way you use other kinds of math. You can apply optimization theory, for example, by noticing that a problem has a certain form, and therefore a certain algorithm will converge to a solution. Applications of category theory are usually more subtle. You’re not likely to quote some theorem from category theory that finishes off a problem the way the selecting an optimization algorithm does.

I had been skeptical of applications of category theory, and to some extent I still am. Many reported applications of category theory aren’t that applied, and they’re not so much applications as post hoc glosses. At the same time, I’ve seen real applications of categories, such as the design of LINQ mentioned above. I’ve been a part of projects where we used category theory to guide mathematical modeling and software development. Category theory can spot inconsistencies and errors similar to the way dimensional analysis does in engineering, or type checking in software development. It can help you ask the right questions. It can guide you to including the right things, and leaving the right things out. [1]

[1] https://www.johndcook.com/blog/applied-category-theory/

2 bulan yang laludiscuss

If you lived in a high place (Denver), you will find it different from a flat lowland (Chicago).

Also in Rio, how high you live can be a marker depending on which part of town you are. Favelas are on hills, whereas wealthy people in Zona Sul live down the hill closer to the beaches.

2 bulan yang laludiscuss

But why though? DuckDB can still be used as a local query engine — I still use it as that. I haven’t touched any of the DuckLake stuff and the duckdb cli and Python library are still my bread and butter. They can add new use cases, but it doesn’t affect the core engine.

Is the concern that the duckdb messaging is now diluted by it having all these extra features? That you can’t sell it to friends as “this thing” like you can a one use tool like curl? I get that, but I also feel that duckdb is so much bigger than a “do one thing and do it well” tool.

It’s an engine that drives the modern data tool stack. Duckdb’s team has been prescient in that it has made many tasteful bets on what users want —- the ability to interop with pandas and polars, addition of geospatial, the plug-in infra. They’re all optional but when you neeed these things, they’re so useful. They’ve also clued me into what the broader data world is thinking about (I didn’t know about sketches and hilbert, but those are so useful in probailistic large scale queries and in geospatial queries). And they exist in larger database systems like Redshift too.

So far duckdb’s bets have been tasteful, and mostly ignorable if you don’t happen to use them.

2 bulan yang laludiscuss

DuckDB is both a standalone and a component. This effort is actually very coherent and brings it back into a familiar usage model — that of a traditional client server RDBMS.

RDBMS have always been multi-user concurrent systems. DuckDB is a very fast local engine that has a multitude of use cases because it is a embeddable in other systems.

It’s like saying what does SQLite wanna be? It’s in your phones, your browser, your desktop apps, iot devices and people have extended it in different directions. The only difference here is that this is first party not third party. But to me it’s a very legible move.

2 bulan yang laludiscuss

When did FT become Business Insider?

I have an FT subscription and they keep moving toward this kind of narrative first reporting to get clicks. It’s no longer a believable paper.

2 bulan yang laludiscuss

It sounds like this replaces the PCA reconstruction function with a quadratic.

The normal PCA encoding:

1) Given a mean-center-scaled X matrix, get the latent variable matrix T with X = T * P’ + e, where P = loadings and e = residuals. The P is your model, so for a new vector xnew, you can calculate tnew = xnew * P (because P’ * P = I).

This is the encoder —- nothing changes here. The original matrix is dimensionally reduced with residuals e discarded. This is why PCA is lossy.

The decoder is where things diverge

The usual PCA decoder reconstructs a given latent variable t_any by using the trained P loadings, like thus x_reconstructed = t_any * P’. This reconstructed data lies on a linear hyperplane, so if the original data did not lie on the hyperplane, reconstruction errors are potetially high.

In your proposal, instead of a linear decoder, you train a quadratic decoder (essentially a classic ridge regression using a quadratic) on the original X. So for your reconstruction, you have x_reconstructed = poly(t_new).

This achieves lower reconstruction error in-sample (naturally, because quadratic is higher order than linear), but your poly function is trained on a particular corpus. Which means that when you’re in-distribution within that corpus, you’re good but when you’re not, you can be very wrong in biased ways that PCA’s linear reconstruction is not.

SO this is not a better technique than PCA in a general sense. It’s a better reconstruction machine when your data is mostly in-sample. It’s a kind of computationally cheap “specialization” on a particular distribution of data, which can be useful if you’re mostly in-distribution but introduces new risks when out-of-distribution.

Whereas PCA just drops the residual and makes modest claims, a quadratic decoder is trying to predict the residual and on out-of-sample data, it can be wrong in biased ways that PCA is not. In other words, it can hallucinate.

But if on a large enough training corpus, chances are we’re going to be in-distribution most of the time, so maybe this could generalize well.

2 bulan yang laludiscuss

Right now Alexa+ and Gemini are objectively better.

The best is ChatGPT voice mode. It understands non English words and accents amazingly well, and even though the LLM model isn’t the full fledged one, I can have deep conversations with it for an hour without it missing a beat.

2 bulan yang laludiscuss

I read this article 10 years ago by a guy named Ricky Yean who went to Stanford as an economically disadvantaged admit and couldn’t shake his poverty mindset and it cost him when he was running a startup.

Why “few successful startup founders grew up desperately poor”

https://rickyyean.com/2016/01/22/privilege-and-inequality-in...

Poverty mindset is maladaptive because it teaches you only money is worth anything, so you hoard it. But in truth time is also worth a lot and sometimes it’s wise to use money to buy time.

3 bulan yang laludiscuss

I feel like 1 is a self correcting problem. If this goes nowhere it will soon be forgotten.

I can think of one example that did go somewhere: Linux.

3 bulan yang laludiscuss

I feel Gastown is an attempt at answering: what if i push the multi-agent paradigm to its chaotic end?

But I think the point that Yegge doesn't address and that I had to discover for myself is: getting many agents working in parallel doing different things -- while cool and exciting (in an anthromorphic way) -- might not actually be solving the right problem. The bottleneck in development isn't workflow orchestration (what Gastown does) -- it's actually problem decomposition.

And Beads doesn't actually handle the decomposed problem well. I thought it did. But all it is is a task-graph system. Each bead is task, and agents can just pick up tasks to work on. That looks a lot like an SDE picking up a JIRA ticket right? But the problem is embedding just enough context in the task that the agent can do it right. But often it doesn't, so the agent has to guess missing context. And it often produces plausible code that is wrong.

Devolving a goal into the smaller slices is really where a lot of difficulty lies. You might say, oh, "I can just tell Claude to write Epics/Stories/Tasks, and it'll figure it out". Right? But without something grounding it like a spec, Claude doesn't do a good job. It won't know exactly how much context to provide to each independent agent.

What I have found useful is spec-driven development, especially of the opinionated variety that Kiro IDE offers. Kiro IDE is a middling Cursor, but an excellent spec generator -- in fact one of the best. It generates 3 specs at 3 levels of abstraction. It generates a Requirements doc in EARS/INCOSE (used at Rolls Royce and Boeing for reducing spec ambiguity), and then generate a Design doc (commonly done at FAANG), and... then generates a Task list, which cross-references the sections of the requirements/design.

This kind of spec hugely limits the degrees of freedom. The Requirements part of the spec actually captures intent, which is key. The Design part mocks interfaces, embeds glossaries, and also embeds PBTs (property-based tests using Hypothesis -- maybe eventually Hegel?) as gating mechanisms to check invariants. The Task list is what Beads is supposed to do -- but Beads can't do a good job because it doesn't have the other two specs.

I've deployed 4 products now using Kiro spec-driven dev (+ Simon Willison's tip "do red/green tdd") and they're running in prod and so far so good. They're pressure-tested using real data.

Spec-driven development isn't perfect but I feel its aim is the correct one -- to capture intentions, to reduce the degrees of freedom, and to constrain agents toward correctness. I tried using Claude Code's /plan mode but it's nowhere as rigorous, and there's still spec drift in the generated code. It doesn't pin down the problem sufficiently.

Gastown/Beads are solutions for workflow orchestration problem (which is exciting for tech bros), but at its core, it's not the most important problem. Problem decomposition is.

Otherwise you're just solving the wrong problem, fast.

3 bulan yang laludiscuss

You're thinking a tax break which is an unconditional subsidy. That relies on the business passing savings through which folks are right to be skeptical about.

But that's not all subsidy mechanisms. The best ones are where pass-through is enforced, not assumed.

You already know of one that works: WIC. It lowers the effective price for customer, which the store receives as reimbursement.

It's not about trickle-down -- that's ideology. It's more about designing the right mechanism.

3 bulan yang laludiscuss

I use DuckDB daily.

In short — It doesn’t crash often at all.

What you may be remembering were reports of exceptional cases where it didn’t handle out of memory errors well. I was one of the people affected. I was running complex analytic queries on 400 GB parquets and I only had 128GB memory. It used jemalloc which didn’t gracefully degrade. They fixed a lot of the OOM issues so it’s more robust now. I haven’t had a crash for a long time.

On normal sized datasets it never crashes.

3 bulan yang laludiscuss

Try DuckLake. They just released a prod version.

You can do read/write of a parquet folder on your local drive, but managed by DuckLake. Supports schema evolution and versioning too.

Basically SQLite for parquet.