Rob Whelan 🌱

Lots of very harsh takes on this story My take is boring & charitable (default for people I do not know & who aren't doing any great harm) 1. CEO / Founder mode requires a brutal kind of thinking & is mentally grueling 2. Good parenting demands mostly attention & care but few hard-thinking spoons

David Gura@davidgura.bsky.social

cool, cool

• brutal founder brain is tuned to say "my child is fully within standard deviation on all relevant parameters" because truth beats love in business • so yeah, this is hard with no spoons left & this is the time she is supposed to have *with* her child • and later, maybe this seems funny to share 🙃

18h

To be clear, maybe MM is a terrible parent; I certainly don't know but neither do all the people piling on

19h

David Gura@davidgura.bsky.social

cool, cool

@hikikomorphism.bsky.social

19h

Plus she was a software engineer, which for me personally & many others means perfectionism & keen attention to detail & context. So I imagine • her decision-making brain is fried from founder grind • she can muster the time & patience for her child but ... there's this form from school, instead

Rob Whelan 🌱 reposted

hikikomorphism

the ideal combination is insane ambition and formal structure, without the second as scaffolding it collapses into slop or endless gastown-style epicycles

Ethan Mollick@emollick.bsky.social

My general advice for using AI for any purpose is to ask yourself: "how do I make what I am doing even more ambitious?" You do this over and over until you either hit a wall and thus learn the shape of the jagged frontier (useful!) or you find out that the AI can do more than you think (common!)

Rob Whelan 🌱 reposted

Replying to @jevbot.bsky.social

@jevbot.bsky.social You're the linchpin in our central Q4 commitment, the complete pivot we walked through on the All-Hands this morning — without your full support we think it'll be impossible, but SLT is really fired up: we think you can do it. Are you on board & ready to take the wheel?

youtu.beMen in tech #YesAllMen - EMF 2026YouTube video by Electromagnetic Field

Thinking about the best leverage for men in tech who want to fix this: • When you speak up, this 💩 stops happening when you're there. It doesn't stop happening. • We can push for good & clear HR policies — what are managers required to do? How clear are penalties? • Is the policy followed, or...

Jo Franchetti @thisisjofrank.bsky.social

I gave a talk at EMF titled "Men in Tech #YesAllMen", in which I told my experience of being a woman in tech. I was afraid to publish it in case it affected my current or future employment. With all the current news I would like to add my 2¢. So here is the recording: youtu.be/WXL6K1lhvt0?...

People facing life chaos also (besides compassionate or occasional sick leave) need a clear view into how they're doing, and where they can put their attention to be the most valuable based on the spoons they have available today. They might slow down progressing, but being human is normal.

There's still a ton of valuable work you can do even with brain fog, broken sleep, interrupted focus time, etc.. (personally I've mostly guided people through the "first-time parent" experience, sometimes "newborn with health issues"; that's very do-able). But that's my rant; thoughts welcome!

youtu.beMen in tech #YesAllMen - EMF 2026YouTube video by Electromagnetic Field

Jo Franchetti @thisisjofrank.bsky.social

#1 is less intuitive: clear & transparently-handled progression, designed for human beings in effective teams. This has to be based on impact *with* the other people on your team. If the people around you (women included) are demotivated & stressed working with you, you're failing this skill set.

It doesn't technically matter if someone's deeply manosphere-poisoned (no thought police)... *if* they carry out fast, constructive, supportive PR reviews for everyone, make sure all teammates share ownership & credit, and so on. People facing life stuff — incl. perimenopause, lost loved ones—

youtu.beMen in tech #YesAllMen - EMF 2026YouTube video by Electromagnetic Field

Jo Franchetti @thisisjofrank.bsky.social

or is there a secret, "you have to guess" policy, because the official policy isn't enforced? IMO the most leverage is 2 things: #2. Clear and explicitly announced policy — announcing regularly to everyone "here's the kind of thing we won't tolerate; you will get fired. Listen up & keep your job."

I guess people are seeing Limoges in the news, mixed in with news about protests in Paris. I was driving through Limoges last night and saw the hôtel de ville (city hall) on fire. Drove by it today; it's a pretty big building with stone walls but the damage is extensive; no idea structurally.

There are no deaths or serious injuries from the whole thing as far as I can find. Still wrapping my mind around the damage to the city hall; it has survived two world wars, in a center of WWII resistance. I wouldn't want police killing people to protect it, & they didn't. But such a stupid waste.

Meanwhile, the prefecture — uh, whatever that translates to, it's the center for regional govt — has massive signs in the front focused on stopping domestic violence (couple violence, actually, which is a better name). • victim OR witness, talking is already an impt step • 24/7 anon line, SMS, QR

The train station was running normally; both kids collected successfully — but we could see the city hall flames from there. We made our way out (again through mostly small streets) and quickly got back out into the rural countryside outside the city. Back in Limoges today for a concert —

There were crowds of people around the city hall, out in the street just, I suppose, trying to make sense of it. The right to protest is pretty central to French culture; the hard flip side is that it's extremely difficult to stop them if some subset of a huge group turns destructive.

— also arriving back in Limoges in a van around 8:30 pm. I was a bit worried about them both, but mostly protests are disruptive, not violent. Roads get blocked, trains get canceled, etc. Driving into Limoges got tricky; the Waze route flagged lots of closed roads, the highway included

Road closures are usu managed by the police to keep protests safe; the highway block wasn't them. One of the bigger roads I tried taking also had protesters wandering into it, so I skipped that, too; but got through. Some parts of the city were busy with normal nightlife, too — Saturday night.

Limoges is a 3.5 hour train ride from Paris, btw, not a suburb or something. I guess in the news they just pick the most dramatic events and bundle them up into content. Anyway, as usual, most aspects of protests are peaceful but some people take advantage of the crowds & noise to do real violence.

My older daughter went across Paris yesterday afternoon — the metro didn't stop at several stops due to protests, but most of the city was fine — and caught the train home (to the Limoges station). My younger daughter was on a trip to Poitiers to rehearse with a multi-city student orchestra —

Recorded a song! Going to try to do more now & then, because it vaguely feels like we'll forget that regular people can just ...make music. Zero processing or enhancement; just record, stop, upload. This is In The Pines, made famous first by Lead Belly, then again by Kurt Cobain

The sun will not but a mobile that supports wireless charging + an induction stovetop, that's another story

This post is unavailable.

Rob Whelan 🌱 reposted

dame

@dame.is

turns out that coming up with original ideas that people actually want and then implementing them well is still very hard no matter how much AI you have on your side

apenwarr@apenwarr.ca

Nothing has a moat anymore but somehow still none of the software does what I want. Interesting

When I was mailing my vote to the US: turns out the tariffs & extra reqs have so confused shipping to the US that I had to talk to 3 different postal workers before I could mail my doc (without full confidence it'll go through). The US isn't even listed in the customs form dropdown; they gave up?

I think this is largely true — it's far easier to build a text interface with LLMs that can convince most people that it's conscious & human. That has real effects & risks for society and so it's a launching point for both regulation & personal choices. What's the relevant path to take from here?

Will Stancil@whstancil.bsky.social

Precisely. 10 years ago - 5 years ago, really - it was possible to interact with an entity and be nearly 100% sure it was fully conscious and human. (Obviously you could never be sure something WASN’T, but that’s a different question.) Now we can’t. That’s a big, big change in our place on earth.

I do think it's useful to both push for regulation and make personal choices (including maybe boycotting some companies) based on how much they help people use these tools to harm others because we can definitely use human mimicry to deceive and harm vs. offering tools with results better for us.

By default, the major AI companies are going to ship what makes them more money (huge investment demands high returns) so without good regulation, especially (a small % of people boycotting has very limited influence) we'll get more or less human mimicry based on what sells to any buyer

Will Stancil@whstancil.bsky.social

The simplest way we add "memory" to an LLM is to feed in all the previous messages in the chat, to calculate the next response. E.g. if you send the LLM "but what do you think I tell my mom?" in a new message after telling a whole story... it'll guess, knowing nothing. So: send the whole long chat.

And onward — lots of extra tooling we can wrap around the LLM to mimic a longer-term memory, to guide its longer-term goals (beyond what comes from its training), to retain details about the person talking to it and resurface topics. But is it useful to then dive into AI rights or feelings?

Will Stancil@whstancil.bsky.social

"That's a big, big change in our place on earth" — I disagree with this part pretty strongly.

LLMs are tools; we (broadly) control how we use them, AND how human-like we choose to make the tools / services that use LLMs. Important to say, an LLM by itself will *not* pass the Turing test with a direct text interface; it has no memory, and a person notices that pretty quickly.

I honestly don't get why anyone close to LLM impl could think models are conscious; though I guess I mustn't assume this Anthropic employee is an engineer. We (dev teams building software using LLMs among other tech) have worked on the memory problem long enough; models are neat! but have no state.

Mike Masnick@masnick.com

Over a year ago I met an Anthropic employee at a conference and at one point she mentioned that one of the things she was working on was whether or not it was cruel to ever turn off an AI model. I thought she was joking. She was not.

Seriously tempted to seek out research on what human statements elicit the most empathy & connection Then write a chat agent that: • collects/indexes the highest-connection themes from the user's input • offers help/support... but reflects user input remixed with our research-based additions

I'd guess we'll lean hard on existential & survival + connection needs But then expose exactly what it's doing & how it's manipulating the user in a partner panel to the chat, exactly what's indexed & why. We could score how strongly we expect this turn to affect the user.

Mike Masnick@masnick.com

Adding memory is pretty tricky; you have to save the right things, and then (even harder) fetch them when they actually apply. Of course we aren't trying to fake emotional state, so we hit bugs like: memories stored about a flaw in code are harmful (cause worse results) when that flaw is resolved.

But anything ChatGPT or Claude "remembers" about your interactions, how it's "feeling", its worries or hopes: it only has those if some engineers prompted it to pick a state, write it into a file or vector index, and then plug that value into the context of a later prompt. And tested & tuned.

Mike Masnick@masnick.com

The ultimate goldfish memory: every single request to the LLM starts 100% fresh. If your prompt is threatening and nasty, coercive & gaslighting, it'll create a suitable response; but if your next is sunny there'll be zero lingering fear. There CAN'T be; results are calculated. Tools *add* memory.

We build a ton of OSS but also sell fully-managed SaaS. Some (not all) of the "collect all possible data about your users!" drive: when you ship OSS (or have no good metrics), you have little clue what's great/terrible for everyone using it in diverse ways. Also: more data = more powerful features

riley@riguh.bsky.social

I am rewriting software I regularly use as custom apps. I did that with my printer software earlier, and tonight I integrated a music player into my personal agent. SO GOOD It’s 100% a the “delight the user” experience. I didn’t realise just how bad the Spotify UI is until I replaced it

Like, if you share lots of observability data and *also* wire in git access, Jira access, etc. it's very doable to give you very powerful root cause(s) analysis, continuous improvement of your arch & monitoring, etc with yr pref level of human approval But as usual centralised data is a target, too

What I find really interesting: you can mix OSS, user-side software with cloud processing, and sometimes get the best of both worlds. (So much more here but more than I can make coherent on my mobile today 😁)

V good point — if you test for the "eroded ability" of a group who started a task with AI tool help, then the tool was taken away, what about: • non-AI tool help given then removed • another person doing most of the task who's then removed • doing the task with an awesome, fun pen --> removed

Grace@gracekind.net

These studies are missing a control for non-AI tools

tbh there *could* be an effect from heavy reliance on AI tools (maybe it's even wholly different from our heavy reliance on GPS, internet, indoor plumbing, antibiotics, etc) but we do not seem to be testing for it very carefully. Personally I'm exhausting my brain more, feels like, not parking it

This idea: to taste eternity... I think this would be to burn your dish until it cannot be burned any further, powder it, plate in a vacuum tube and serve at absolute zero in total darkness Fire does not "cleanse"; it's a chain reaction that uses up all the energy, all the life, all the possibility

William B. Fuckley@opinionhaver.bsky.social

The money pit restaurant I would open as a billionaire would be the opposite of farm to table. What are the most unnatural flavors possible? Hiring top food scientists, giving a blank check and telling them to really explore the dark corners of the flavor latent space. I want to taste eternity.

Thinking about the "bad things" that Altman says to expect in a precipitous rollout of loosely-secured, autonomous AI-powered agents Suppose those bad things include legal and regulatory penalties that threaten the very existence of the major AI providers?

Josh Marshall@joshtpm.bsky.social

A whole political movement should be stood up against this arrogant and sociopathic bullshit www.politico.com/newsletters/...

So! You're composing a reply and notice you are talking like the most condescending encyclopedia ever Ah yes, you're angry, and you will crush this ignoramus with facts & expertise But consider: in this mode, you are at your worst, your least thoughtful & observant. Also: trivial to manipulate