HNHacker News
TopNewBestAskShowJobs

blueboo

793 karma · joined May 9, 2018

submissionscomments
blueboo··on Launch HN: Martin (YC S23) – Using LLMs to Make a Better Siri
The TiVo problem https://a16z.com/distribution-vs-innovation/
blueboo··on Show HN: Wat – Deep inspection of Python objects
rich.inspect has been my go-to for this. Comes pre-installed in a vanilla Colab kernel too
blueboo··on Why don't they compose music like Bach any more?
Just because there wasn't Twitter and App Stores doesn't mean there weren't fashions and markets -- they were located in different places in the culture. I don't mean to deflate your lofty romanticizations of the man and his craft. Take another listen to your documentary and observe how his career reflects and reacts to the on-going musical-cultural discourse of his time.

In that way, he truly is like any other artist (so engaged.) At the same time, rest assured we can celebrate and enjoy the extraordinary achievements made in that paradigm.

blueboo··on What would it take to recreate Bell Labs?
It rings a bit false when juxtaposed with their $40b spend on the Metaverse…where was the savvy leveraging of the open source community then?

Meanwhile Meta’s competitors commoditise and glean profits from actually-SOTA LLM offerings.

In any case, their hypothesis is testable: which open source innovations from Llama1/2 informed Llama3?

blueboo··on Why don't they compose music like Bach any more?
Bach was a fashion(-driven) designer like any other artist, reacting to and driving trends of his time.

With that framing, they never stopped composing like Bach

blueboo··on What would it take to recreate Bell Labs?
Demis on designing DeepMind:

"When we set up DeepMind, I took inspiration for our research culture from many innovative organizations, including Bell Labs and the Apollo program, but also creative cultures like Pixar."

https://www.theatlantic.com/sponsored/google-2023/unlocking-...

As for materials science, there's signs of life in graph nets https://deepmind.google/discover/blog/millions-of-new-materi.... Meanwhile maybe AlphaFold indicates a precedent towards deep learning systems that advance chemistry. I don't claim these are revolutions, but extrapolating into the future, they might be seen as the catalysts. Maybe.

blueboo··on Show HN: I made a looper for YouTube clips
Is this better than looptube.xyz , YouXube.io, LooperTube, RepeatTube, YouTubeLoop or any of the other hits for “loop youtube”?
blueboo··on Apple okays Epic Games marketplace app in Europe
Ah but if they’re forced to compete, the same “crap” might not automatically top the charts.
blueboo··on My finetuned models beat OpenAI's GPT-4
Why would you set temperature=1 for this task?
blueboo··on Show HN: I created an After Effects alternative
What would it take to support AE plugins? Nail that and you have something to make Adobe nervous.
blueboo··on Anybody have problems with bookmark organization like me?
There are many tools. Save-and-triage with Todoist, tag and search with Pocket, knowledge management tools galore.

I’m going to strawman a bit here, this might not be your issue, but I think it’s useful because this is a common pathology. The tool isn’t the problem. The problem is the underlying FOMO. You can’t read everything. You’re not going to read everything even if you bookmark it in the perfect app.

What if, instead of bookmarking, you scheduled time in your calendar for reading it to the extent you felt you’d extracted optimal value from that resource? How many hours a week would you dedicate? How many pages would that imply?—and how does that compare with your bookmarking appetite?

Anyway the better thing was here all along, it’s just only recently even enabled by multimodal retrieval: screenshots

blueboo··on Language is primarily a tool for communication rather than thought [pdf]
I find this unconvincing. Words described those terms just fine as you just proved.
blueboo··on Safe Superintelligence Inc.
it's intellectual ego catnip. https://idlewords.com/talks/superintelligence.htm
blueboo··on Google DeepMind shifts from research lab to AI product factory
I dunno. Android, Chrome, ChromeOS, Workspace. There’s a lot there that’s knit together. Compared to what, Safari, Pages and Messages? Teams, Word, and Outlook?

They all have their issues. Maybe the crux of it is chat. If your work is integrated into your chat — the day to day online social space — it’ll feel “integrated”

blueboo··on Tesla's FSD – A Useless Technology Demo
Fake it til you make it is a fundamental principle of startups. We just don’t usually see it at such a vast scale.

There’s a timeline where Theranos was acquired for 9b by UnitedHealth if they could keep the grift alive juuust a bit longer and Elizabeth Holmes ascends to the tech firmament permanently while her enablers congratulate each other.

Tesla has even more and deeper financial and branding defense mechanisms. That said, the clock is ticking, now, I think

blueboo··on U.S. clears way for antitrust inquiries of Nvidia, Microsoft and OpenAI
Their competitors didn’t win. That’s not the same as failing, insofar as building technology goes. Business, sure
blueboo··on Mp3tag – Universal Tag Editor
Love mp3tag. And now in 2024 we can go one step further -- LLMs are the ultimate solution to finally tagging your entire collection consistently.
blueboo··on Shane Legg: Machine Super Intelligence (2008) [pdf]
even AGI has its limits.
blueboo··on Design docs at Google (2020)
Design docs are vehicles for discussion. The idea is that they are the most efficient way to communicate your intention, motivation, why not alternatives.

They are one mechanism that fits in a broader working culture. If you’re working solo, it’s an extravagant exercise. If you’re in a huge team, it’ll enable leveraging more expertise across the team and act as documentation.

There are a few different failure modes.

* Valuing output not outcomes is classic misalignment. (writing 40 page docs for promo — this doesn’t work except for very junior folks who are proving the ability to string words together more than deep eng)

* As mentioned, it’s a bit much of solo teams. Other small teams can comfortably communicate through issues (Jira) plus breakout sessions to hash out ideas.

* Engineers also need to be onboarded to effective design doc writing. (note a top commenter frustrated that their first effort isn’t immediately met with adulation, a possible flag.)

Writing about code is hard. Usually this is an exercise celebrated on HN. If you’re in a team, beware if you find your work is always that which doesn’t need to be explained and deeply considered in a shareable doc.

blueboo··on Ask HN: Do startups building on OpenAI have a defensible moat?
They’re a PM’s whim away from being steamrolled by their foundation model vendor. And these startups are doing the product market fit for them

The bigger the opportunity, the more general the audience, the juicier the B2B market, the greater the risk.

Time was when Jasper was valued more highly than OpenAI. Now it’s flagging and OpenAI is headed for 12 digits…

blueboo··on MIT abandons requirement of DEI statements for hiring and promotions
Well, the “acknowledge and mark” phrase gets at the statement—That if the org thought objective X was authentic existential goal, it behooves them to understand how an applicant understands and would configure into that plan. (I have no claim about its efficacy. The highly-structured nature of the statement is a head-scratcher. I’m here watching the fallout with everyone else.)

I hasten to restate that is my understanding of the premise, in the spirit of collectively untangling the causal chain here. This is incendiary stuff on HN!

blueboo··on MIT abandons requirement of DEI statements for hiring and promotions
This is the premise:

There are outstanding systemic inequities at every level of American society. The threat these represent to our long term survival and prosperity are such that we are keen to at least acknowledge and mark our and our prospective collaborators’ efforts to improve the status quo.

Maybe you don’t accept that. Maybe it’s asking for a virtue signal in service of performative posturing. Maybe this requirement had adverse effects (evidently it has.)

But there is a through-line of coherent logic, and the total failure here may be cause for alarm.

blueboo··on They thought they were joining an accelerator – instead they lost their startups
the stupendous risk taken on by founders confounds instincts, and we're already filtering to folks who have resorted to taking money from third- or fourth-tier options
blueboo··on GitHub Copilot Workspace: Technical Preview
First of all, yes, this is a provocative prompt that bears engagement. You're right to be concerned.

I share your frustration with the reticence of seasoned engineers to engage with these tools.

However, "structuring and executing large-scale changes in entire repositories" is not a capability that is routinely proven out, even with SOTA models in hellaciously wasteful agentic workflows. I only offer a modest moderation. They'll get there, some time between next week and 2030.

Consider: Some of the most effective engineers of today cut their teeth writing assembly, fighting through strange dialects of C or otherwise throwing themselves against what are now incontestibly obselete technologies but otherwise honed their engineering skills to a much higher degree than their comrades who glided in on Java's wing.

Observe that months of hand-sculpted assembly has turned into a single Python call. AI is yet another tier of abstraction.

Another lens is application -- AI for X domain, for X group, for X age, for X culture. Lots to do there.

Finally, there's empowerment. If this technology is so powerful, do you concede that power to others? Or are you going to be a part of the group that ensures it benefits all?

FYI, OpenAI published a labor market study suggesting professions that are more or less exposed to AI. Take a look.

blueboo··on Meta does everything OpenAI should be
Also is perfectly aligned with Yann’s goals as an (academic) researcher whose career is built on academic community kudos far more than, say, building a successful business
blueboo··on Reverse engineering a software crack
Quite convenient that you’d only pay software if developers contacted you years in advance, and none of them were clever enough to do so
blueboo··on WD launches 4TB SD card
Agreed, it’s truly mind boggling. My moment: coming home from a computer show with a spectacular deal — $80 for an 80Mb drive to be installed in our Compaq 286 suitcase unit. (Imagine having 100mb!)
blueboo··on Ask HN: Is programming a drug?
Brooks in MMM identifies five reasons people love programming.

“1. The sheer joy of making things.

2. The pleasure of making things that are useful to other people.

3. The fascination of fashioning complex puzzle-like objects of interlocking moving parts, and watching them work in subtle cycles, playing out the consequences of principles built in from the beginning.

4. The joy of always learning, which springs from the nonrepeating nature of the task.

5. The delight of working in such a tractable medium. The programmer, like the poet, works only slightly removed from pure thought-stuff. He builds his castles in the air, from air, creating by exertion of imagination.”

There’s real pleasure to building software, as long as you stay above drudgery, churn, and/or repetition

blueboo··on Udio: Generate music in your favorite styles with a text prompt
Natural language is how we talk about music and we’ve been talking about music since time immemorial.

What’s your favourite music? Why?

blueboo··on Show HN: Sonauto – A more controllable AI music creator
If there’s variance in output, it stands to reason you’d generate many X your desired output count and curate. Standard practice for creative output, from Midjourney to LLMs
← PreviousPage 4 of 20Next →