Will Kramer

CEO of @laila_team | Coder and AI whisperer

[Will Kramer @willkramer]

The SPV workaround has let companies stay private for way too long. Counting every investor behind the SPV toward the shareholder cap would push these companies public earlier and give retail investors a chance to buy in at realistic valuations.

Amazon went public in 1997 while still losing money and the public markets handled it fine. $1K in $AMZN at IPO would be worth ~$2M today. Secondary markets need to go away too. If you want liquidity for your employees, then go public 🤯 Stripe shouldn't still be private and if the SPV rule were changed, I bet they wouldn't be.

[Will Kramer @willkramer]

@alexsllater Great launch video and slick UI. But I have a hard time believing you paid the roughly $50k it would likely cost to license a Billy Joel song.

Playing fast and loose doesn’t exactly inspire confidence when you’re also asking us to trust you with our personal info.

[Will Kramer @willkramer]

@ArcanesValor @signulll Are you forgetting about Google? They’re not just a frontier lab. They also design their own chips (TPUs) for their own hyperscale data centers, and they have direct access to sources of raw training data. It doesn’t get much more vertically integrated than that.

[Will Kramer @willkramer]

@0xr3dd0g @FletchMatlock Don’t most traditional political ads use their opponents’ likeness without their permission? You know the ads: the scary music with their opponent’s photo.

[Will Kramer @willkramer]

@PalmerLuckey @JoshLeibner Instead of banning anything, USPS should raise the price of bulk postage to fully cover delivery costs. That way there are no subsidies and lower-quality junk mail will naturally decline.

[Will Kramer @willkramer]

Because ChatGPT didn't call a tool. It should have used a randint() call but instead just predicted the next likely token from its training data and humans are heavily biased toward "feelings." Odd numbers, primes, and numbers ending in 3 or 7 "feel" random.
https://t.co/YiOjnVn3Yd
and
https://t.co/GNUc24NfR4

[Will Kramer @willkramer]

@mreiffy @RandianCapital What do you expect when you cede accountability? Evan holds Class C stock with 10 votes per share. Early investors got a Class B which gets 1 vote. And the Class A stock the public owns? Zero votes.

I would never buy $SNAP for that reason alone.

[Will Kramer @willkramer]

@flippatrak @eglyman I think where costs are headed, especially once VC subsidies dry up, will force this shift. Take the model Ramp was promoting: a 0.35B parameter model running at full precision probably uses under 2GB of RAM, which means practically any computer can run it locally.

[Will Kramer @willkramer]

@rohanpaul_ai I get that PayPal is bloated but who would ever broadcast layoffs over the next 2-3 years? Every decent employee will start job searching. What an unforced error.

[Will Kramer @willkramer]

@bradgessler The more I use Azure and GCP the more I miss AWS's actual console.

[Will Kramer @willkramer]

@jasonlk Disagree. "No pure managers" treats AI as if it dissolves the management work. It doesn't. AI reshapes IC output. It doesn't coach, mentor, navigate politics, or make the judgment call on a struggling teammate. AI can run a playbook, not people. https://t.co/lSsTq1vsVi

[Will Kramer @willkramer]

@GergelyOrosz Was an early Copilot Enterprise adopter. $MSFT made you pay upfront for a full year, no trial 🚩

Supposed to work with Excel, PowerPoint, Word. Literally was useless. No one used it. Claude Cowork can actually make Excel and Word docs. Copilot can't...

[Will Kramer @willkramer]

@bhalligan @nifinet That's interesting. I'd flip it so the rep has to reject the lead within X days, otherwise they get charged by default. Nothing worse than a hardworking SDR (your agent) paired with a lazy rep who lets leads go stale and then blames the SDR.

[Will Kramer @willkramer]

@parkerconrad Rippling covers maybe a third of the SOC 2 control surface at most. It doesn't touch your cloud, your code, your SIEM, or your incident response. For that part of the audit, aren't you "purely a detection system" too? What makes this different/better?

[Will Kramer @willkramer]

@systemdesignone Hardest to write: Assembly
Hardest to master: C++

[Will Kramer @willkramer]

@josh_uglyasf Not really a fair question. Hindsight eliminates the risk.
A better one: 2024 with 500k or 2051 with 1M?

[Will Kramer @willkramer]

@signulll Momentous Whey Protein. Made from 100% European grass-fed cows.

[Will Kramer @willkramer]

@systemdesignone Definitely a jerk response and it proves the candidate made the right call dropping out. Reflects badly on the hiring manager. He should have skipped the snark and just said "Thanks for letting us know."

[Will Kramer @willkramer]

This is a strong take on the Cursor/SpaceX deal 👇

[FleetingBits @fleetingbits]

some quick thoughts on the xai cursor deal

1) spacex has agreed to either pay cursor $10bn for some joint model work or otherwise to acquire cursor outright for $60bn later this year

2) i suspect that elon thinks of the $50bn option to buy cursor more as a performance incentive than as a strict option in some financial sense

3) the deal solves a couple of things for spacex

3a) first, it gives them a team that can train near-frontier coding models and hopefully train frontier models with access to xai compute

3b) second, it gives them distribution, it doesn't seem that xai ever had a real customer base outside of the elon verse

3c) third, code is probably important for recursive self-improvement and it seems that code performance is important for the next stage of model development

4) with respect to cursor's data, i'm not sure how important it is; on one hand, i would guess about 30% of cursor's users let cursor use their data, so there are a lot of traces

5) but business users, which probably have the most valuable traces, do not let cursor use their data and anthropic, which has the least aggressive user data terms, does not seem slowed down by this

6) for cursor, i think this is a good outcome, cursor is dependent on frontier labs giving them their best models for sale

7) but those frontier labs on the long run don't want cursor between them and their customers; and will look to figure out how to disintermediate them

8) so cashing out now, at a time when they still have very significant distribution and $2bn arr, and after their team has proved they can train near-frontier models, seems good

9) for spacex though, one issue has been that elon's management style has not been effective with researchers; it seems that he tends to treat them like they do not have a better option

10) for engineers that want to work at spacex or tesla; there is no other place that they can go in the united states to work on space or electric cars at the same level

11) but researchers always have better options, they can go to openai, deepmind, anthropic or otherwise join or start a neolab and this means they need to be managed differently

12) and so, even if the acquisition goes though, there is still a risk that elon will have difficulty managing the acquired talent to continue to produce subsequent frontier models

13) but, i expect him to continue, like zuckerberg, to try things until he finds something that works, at least for the foreseeable future