Dima

Steak holder, stuff engineer. Building @xmemory_ai. Host of the @SysDesignMeetup. CS generalist, ex. competitive programmer.

[Dima @UniqueDima]

Hi @WisprFlow!

I'm a happy paying user on my MacBook, and I'd love to use Wispr Flow on my iPhone too. But my account with you is tied to my Gmail that I do not want to sign into from my iPhone.

I'm trying to log into Wispr Flow with email and password, but as I type in my Gmail address, it required me to do Google sign-in on the phone — which I'm reluctant to.

Any way to work around this? Send an email login code? Generate a QR code in the desktop app? Create a WisprFlow password for my SSO-based Gmail login?

Or perhaps your support could just move my paid subscription to another Google account of yours truly altogether? Happy to converse in DMs.

Would much appreciate your help, and thanks in advance!

[Dima @UniqueDima]

If the self-proclaimed Beats were condescendingly called beatniks because the Sputnik Crisis catalyzed the movement, and given Gen Z is about as f’d economically as the New Golden Age is not there yet, I humbly suggest BeatX as the name for the renaissance.

[Dima @UniqueDima]

[1] https://t.co/2l3auDs6ox

[Dima @UniqueDima]

𝗛𝗼𝘄 𝘁𝗼 𝗺𝗮𝗸𝗲 𝗔𝗜 𝗿𝗲𝗺𝗲𝗺𝗯𝗲𝗿 𝗰𝗼𝗿𝗿𝗲𝗰𝘁𝗹𝘆, 𝗯𝗲𝘆𝗼𝗻𝗱 𝗥𝗔𝗚𝘀 𝗮𝗻𝗱 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 𝘄𝗶𝗻𝗱𝗼𝘄𝘀.

Let's chat, live, in San Francisco.

Tue, Jun 2, 11:30am.

At the Snowflake conference, Braindate Lounge — Basecamp, South Hall A.

Link in [1].

[Son Luong @sluongng]

Codex just found a “workaround” of not having sudo on my pc…

[Dima @UniqueDima]

This AI-assisted coding boom will teach many if not most engineers that being a tech lead is too damn hard.

On the other hand, engineering teams used to produce code much slower than modern-day swarms of agents.

I have a vague idea that the future is some pre-trained / pre-prompted "chat windows", or "ready-to-join conference rooms", where our freshly minted human tech leads can chat or talk to the AI about the vey code that needs to be understood and evolved further.

So that the human part of the review process will mostly be some "Do I understand it correctly that ...?" questions, with various agents working together to produce meaningful answers from different angles.

And we will need harness for this tooling too. What a time.

[Dima @UniqueDima]

After some basic cleanup, the largest two files on my MacBook are ~/𝙻𝚒𝚋𝚛𝚊𝚛𝚢/𝙰𝚙𝚙𝚕𝚒𝚌𝚊𝚝𝚒𝚘𝚗 𝚂𝚞𝚙𝚙𝚘𝚛𝚝/𝙲𝚞𝚛𝚜𝚘𝚛/𝚄𝚜𝚎𝚛/𝚐𝚕𝚘𝚋𝚊𝚕𝚂𝚝𝚘𝚛𝚊𝚐𝚎/𝚜𝚝𝚊𝚝𝚎.𝚟𝚜𝚌𝚍𝚋 and ~/𝙻𝚒𝚋𝚛𝚊𝚛𝚢/𝙰𝚙𝚙𝚕𝚒𝚌𝚊𝚝𝚒𝚘𝚗 𝚂𝚞𝚙𝚙𝚘𝚛𝚝/𝙲𝚕𝚊𝚞𝚍𝚎/𝚟𝚖_𝚋𝚞𝚗𝚍𝚕𝚎𝚜/𝚌𝚕𝚊𝚞𝚍𝚎𝚟𝚖.𝚋𝚞𝚗𝚍𝚕𝚎, in this order.

Each one almost 8GB.

[Dima @UniqueDima]

Obviously, it’s time for more people in the world to learn that payments in the form of payments are something the other party can intercept.

Without taking sides here, paying one’s legal fees and/or offering them goods and services in a different state is clearly the next step in this escalation strategy.

I, for once, would love to see my country find more effective ways to fight measures as outrageous as what California attempts to introduce.

Because we need both the people who are willing to resist those measures, and the history of precedents on how to fight back effectively.

(Spoken as a WA resident who’s unhappy about WA introducing an illegal income tax here.)

[Governor Newsom Press Office @GovPressOffice]

BREAKING: California will impose a 100% tax on payments distributed from Trump’s Jan. 6th “slush fund.”

[Dima @UniqueDima]

Keep talking about this!

[Dreams of Code @dreamsofcode_io]

Really good time to consider putting your SSH Keys on a hardware security key, such as a Yubikey.

[Dima @UniqueDima]

𝐈𝐦𝐩𝐨𝐫𝐭𝐚𝐧𝐭 𝐪𝐮𝐞𝐬𝐭𝐢𝐨𝐧: 𝐢𝐬 𝐢𝐭 𝐭𝐫𝐮𝐞 𝐭𝐡𝐚𝐭 𝐋𝐋𝐌 𝐦𝐨𝐝𝐞𝐥𝐬 𝐚𝐫𝐞 𝐛𝐞𝐭𝐭𝐞𝐫 𝐭𝐡𝐚𝐧 𝐨𝐭𝐡𝐞𝐫 𝐋𝐋𝐌 𝐦𝐨𝐝𝐞𝐥𝐬 𝐰𝐡𝐞𝐧 𝐢𝐭 𝐜𝐨𝐦𝐞𝐬 𝐭𝐨 𝐬𝐮𝐠𝐠𝐞𝐬𝐭𝐢𝐧𝐠 𝐡𝐨𝐰 𝐭𝐨 𝐨𝐩𝐭𝐢𝐦𝐢𝐳𝐞 𝐩𝐫𝐨𝐦𝐩𝐭𝐬 𝐟𝐨𝐫 𝐭𝐡𝐞𝐦𝐬𝐞𝐥𝐯𝐞𝐬?

Say I have a process which involves querying the model again and again. And as the good ML/AI citizen I am, I have labeled data — journaled and marked-down records of where the model performed well, and where it might need improvement.

Fine-tuning and post-training the model are of course plausible directions. But that's slow and expensive. A much cheaper alternative is tweaking the prompts.

So I have the means to incrementally run some end-to-end test. Or, in pure SGD terms, I have a process to incrementally approximate some gradient stochastically.

As the result of this process I get a suggestion on what can be improved in the prompts, so that the result gets Pareto-better. Presumably. So: check and repeat.

I also use other models to cross-check those improvement suggestions, to make sure they don't overfit to the very problem — since it's often generalizing a particular failure mode from just a few examples.

This is plausible. This works. This is cost-effective.

But one thing doesn't let me sleep well at night. We have this belief that 𝐦𝐨𝐝𝐞𝐥 𝐀 𝐢𝐬 𝐛𝐞𝐭𝐭𝐞𝐫 𝐚𝐭 𝐢𝐦𝐩𝐫𝐨𝐯𝐢𝐧𝐠 𝐩𝐫𝐨𝐦𝐩𝐭𝐬 𝐟𝐨𝐫 𝐦𝐨𝐝𝐞𝐥 𝐀. What is this belief based on? Is it even true?

It sounds wise to use Anthropic models to improve prompts that are later fed to Anthropic models. Replace "Anthropic" with any LLM provider here. But do we know whether it's true at all?

Perhaps the next big thing is some Grok training a model that can prompt some Qwen better? Asking for a friend, of course.

Seen in this light, all in all, the idea of a proper bench of models and model ensembles in a closed-loop system starts to look more and more lucrative.

[Dima @UniqueDima]

TIL it's not only easy but encouraged to use the CLI agent from @cursor_ai for CI gates!

Which is great, because I'm very paranoid when it comes to Anthropic and OpenAI keys. They can sign me out and then it's pain to rotate the token. They can lock me out since it's a gray area in terms of the EULA. And they can cost me a fortune.

The safest option I found so far is OpenCode with browser-based $20 ChatGPT Plus auth. But that runs out quickly, both the 5-hours and the 7-days windows.

And here I am, once again happy about paying for Cursor since forever. Because that'd be my solution of choice now.

Right, @cursor_ai folks? I can totally use my API key to power LLM-based Github CI gates for our small team — right?

[Dima @UniqueDima]

The Short Interval Between Two Eras

Here's the post I forgot to write a few weeks ago, right after watching Project Hail Mary.

(The book is excellent too.)

One distinct aftertaste of the movie is how precisely the short interval between the book and the screenplay coincided with the short interval between the pre-AI and post-AI adoption era.

Just a few years ago, computer interfaces were presumed to be incredibly difficult to manage. And computer programming — a.k.a. software engineering — was widely understood to be a difficult discipline.

Andy Weir is famously a "science maximalist": his books are ripe with the idea that The Science has solutions to all problems. Yet, in The Martian, I cannot help but wonder how it is possible that a botanist navigates the intricacies of a complex software system so flawlessly.

Reading Project Hail Mary, I found myself thinking along the same lines. Being good at first-principles physics does not automatically make one proficient with bleeding-edge tech — which, inevitably, is what an interstellar spacecraft is full of.

And yet! By the time the movie is released, we know for a fact how powerful human-first, natural language interfaces can be.

Quite literally just five years ago, the thought that an interface could be both powerful and intuitive was unimaginable. Sure, sci-fi authors had been talking about this possibility forever. But the tech community — yours truly included — was rather sceptical.

AI-assisted coding has changed this in a matter of single-digit years. An astonishing product progress, if you ask me.

Moreover, Neuralink et al. is no longer something impossible. Muscle memory is tricky — it'll be a while until one can wake up like Neo in The Matrix, knowing Kung Fu after a short session.

But for intellectual tasks — for virtually all of them — the problem can largely be declared solved. To my taste, if one can understand the domain, articulate the desired outcome, and answer a few clarifying questions, then even today's AI models are quite capable of making things happen at astronomical scale. Literally astronomical: tasks such as planning extraterrestrial space travel are a piece of cake for a mid-sized AI model equipped with just a few tools.

What a time to be alive (c)

[James Woods @RealJamesWoods]

Greatest one-line tweet ever.

[Dima @UniqueDima]

AI memory is rocket science, they say.

Build the best product and win they say.

It's not a hype-y market it's a real deal they say.

/me still optimistic — @xmemory_ai is the best memory solution on the market, although it might take time time for the market to realize that.

"If those choosing memory solutions could read the benchmarks and whitepapers, they'd be real upset." (c)

[Dhravya Shah @DhravyaShah]

"Can I copy your homework?"

"Yeah just change it up a bit so it doesn't look obvious you copied"

"Ok"

[Dima @UniqueDima]

As we've definitely entered the era of smart AI agents, I think it's time we go back to the roots with shell scripting.

Say no to long shell scripts. Say no to complex business logic in shell scripts. Instead, this logic belongs in plain English!

Replace one script with ten options by ten shallow do-one-thing-and-do-it-well scripts. Or fifteen, if some of them have modes and start / stop / check functionality.

Then write a plain English file, likely markdown, outlining for the agent what these scripts are, how to use them, and what to keep in mind. A human may also read this English file, but who are we kidding?

And then invoke these scripts via a decent AI agent, such as Cursor. In your workflow, as needed, it will present a concise summary, with tables where needed.

If uncertain about the output, just ask it again, in plain English. And if some script output needs proper visualization, that's definitely not the job of the script itself, but of its outer harness.

And this outer harness is the human operator and an AI model. A model so small that it'll be running locally in just one or two generations of our laptops.

What a time to be alive.

[Dima @UniqueDima]

With the increasing frequency of supply chain attacks, perhaps the next big hardware product is some Raspberry Pi with Ubuntu or Alpine, with a physical button to reset it to its vanilla state in a few seconds.

Then have it seed its config with SSH auth and some metered LLM models access from a QR code with my public key.

Better yet, a "Morse code seed", which it hears, since microphones are cheaper than cameras. Although or NFC via some YubiKey — secure and quick.

I'd literally buy a "six-pack" of those, just to use them in a fungible fashion, and to rotate them often.

[Dima @UniqueDima]

If LA has discontinued law enforcement practices that disproportionately affect some social groups, any and all tax increases should also be banned as they disproportionately affect some other social groups — what am I missing?

[Dima @UniqueDima]

In a sane ecosystem a competitor to Google Maps reviews on which can’t be silenced emerges overnight, and takes over by a storm. I’m sure @levelsio would take single-digit hours to build a usable one.

And our blockchains are good enough these days. For an anonymous solution, cash discounts, etc.

And clearly, the customers would benefit. And good restaurants would win as well.

I struggle to find a single non-moronic reason to design the regulatory framework this way. And yet here we are.

Pathetic Europe.

[@levelsio @levelsio]

My gf is banned from reviewing places in Europe on Google Maps after she gave one restaurant in Portugal a 1-star review

When she reviews inside EU it gets auto rejected, outside EU she can review any place

Free speech in Europe has sadly died a long time ago

[Dima @UniqueDima]

Weird how Elon says he [co-]founded OpenAI because he wanted a counter to Google’s approach to AI: closed, private, for-profit.

And as of now the largest benefactor of the ongoing trial is indeed Google — which did soften up since, but all of the original arguments against it by Musk clearly hold true.

[Dima @UniqueDima]

Speaking of losing faith in your heroes, and speaking of the public and the consumer being largely responsible for the BS happening around us.

This: https://t.co/OkXg99yLrS

A tragedy if you ask me. And NOT A SINGLE mention of this by Jeff Dean himself, or when you search for his name. You literally have to add Berkeley to get the “original” coverage.

It’s not Berkeley who are the jerks here. It’s the people who refuse to speak up about the unacceptability of this very selective law enforcement. And the latter group, of those silently complicit, includes most of the Bay Area.

How this is not all over the news is beyond me. How Google is not announcing that from now on it is sending its own security to ensure students can enjoy the event they are coming to enjoy is beyond me. How the provost and the administration of tue university still keep their well-paid positions is beyond me.

And this is a rather benign example tbh. There exist far, far worse cases.