Eliot Payne

Smile!

Later today you won't feel like it.

[Nscale @nscale]

Heading to @RaiseSummit? The Nscale team will be at the Carrousel du Louvre from Wednesday.

MD Kristin Zwez will also be joined by representatives from @NVIDIA and @VAST_Data as she moderates a discussion on time to first token and how the full infrastructure layer is now building to improve it.

📅 July 8 - 2:00PM | 📍 Master Stage

[Alex Payne @AlexPayne]

Should have had AI proof-read my slides!

[Eliot Payne @eliot_payne]

@alexpayne Wow very cool bro!

[Alex Payne @AlexPayne]

I’m not getting any notifications because X is down right?

[Alex Payne @AlexPayne]

try this in operator to find bugs and create issues in github whilst you sit back and relax:

Prompt: First log into github and check existing issues for <<repo name>> . Then Log in to << your project url>> and identify 10 new areas or items of the application that are not complete. List what they are, where they are located, and what still needs to be completed. Then, log back into GitHub and create issues for each of these incomplete areas in the repository for better tracking.

Important: github contains VERY sensitive data so you need to watch operator carefully - there are already some safeguards in place for this, e.g if you go to another tab the task will automatically pause.

I like this task because it automates something that I personally find very boring and mundane to do, watching operator complete this is very satisfying!

[Alex Payne @AlexPayne]

Prompting tips for Codex:

Just like ChatGPT, Codex is only as effective as the instructions you give it. Use the following tips:

1) Use greppable names. Codex literally calls grep, so specific filenames, symbols, or unique package names help it quickly find the right spot. Internally, OpenAI use the prefix 'wham' for Codex-related packages.

2) Tell it where to work. Codex performs best when pointed at a single file or at most a package containing ~100 files. Broad or vague prompts can leave the agent guessing.

3) Paste the full stack trace. Exact stack traces with file paths and line numbers help Codex immediately pinpoint bugs.

4) Spin up multiple tasks in a row. Each task runs in its own isolated environment, so feel free to queue multiple tasks simultaneously. Many engineers at OpenAI start their day by making a quick to-do list and firing off several tasks at once.

5) Give it work with a pass/fail. Just like a human, Codex validates its changes. Since it has access to a terminal, anything verifiable with a unit test or linting lands more reliably. (Codex doesn't yet support UI tests.)

6) Split large changes. Instead of giving Codex a giant PR, break work into small, focused tasks. Smaller tasks are easier for the agent to test individually and easier for you to review.

7) Let Codex take over when you're stuck. If you get blocked, create a branch and hand the problem to Codex. You can use this strategy to explore multiple solutions in parallel.

8) Kick off a few tasks before you start your day. Launch tasks before your commute or morning coffee, and come back to fresh diffs ready for review.

These tips and full docs can be found 👇

[Alex Payne @AlexPayne]

o3 pro today?

[Alex Payne @AlexPayne]

here’s a quick and dirty tour of the closest historical parallels to an AGI driven future

(because let’s face it, no one’s really ready for it)

1. industrial revolution (1760-1840)

new machines came in hot and skilled workers freaked out (luddites smashing looms, staging riots, pretty intense)

short term it sucked for a lot of people long term new jobs popped up and average wages actually rose (though unevenly, and slooowly)

lesson? tech shocks hurt in the moment but new industries usually follow, IF we invest and actually share the $$$ gains

2. gilded age & second industrial revolution (1870-1914)

railroads, oil, electricity = massive $$$ for robber barons, everyone else? not so much

strikes got violent af (pullman 1894, homestead 1892)
but that chaos forced new laws: anti-trust, income tax, social insurance, universal schooling

lesson? policy matters. tech concentrates $$$ fast, so you better balance it out politically with taxes, education and collective bargaining (or shit hits the fan)

3. mechanisation of agriculture (1900-1960)

farming jobs tanked from ~40% to <5% in the US over six decades as tractors replaced workers

but social meltdown never really happened. people just moved to cities and found new gigs in factories/services
(also ww2 helped soak up all those workers)

lesson? speed matters. slow tech changes = easier transitions, fewer riots

4. mid-20th century office & factory automation

atms, mainframes and factory bots sparked fears of mass unemployment (“cybernation” was the og agi scare)

didn’t happen tho

(atms actually made banks cheaper so more branches opened, meaning MORE tellers overall. go figure)

lesson? automation often remixes jobs rather than killing them outright, and cheaper services can actually boost demand

5. IT & globalisation (1980-now)

pc, internet, offshoring = hollowed out middle jobs, gig economy, wage gaps

people didn’t riot much tho, they just voted for populists (brexit, trump) new roles popped up too (app devs, data scientists, ecommerce logistics)

lesson? cognitive automation shakes shit up big time too

but how bad it gets depends on retraining, safety nets and worker power

6. how agi might be different (spoiler: it’s scarier)

past tech waves: mostly physical/routine cognitive tasks, slow rollouts, high capital cost, humans still needed

agi wave: everything digital could be automated (even high-status knowledge jobs), rollouts in months not decades, super low costs (cloud APIs anyone?), human designers optional

(literally NO historical precedent)

lesson? agi might compress centuries of disruption into a single business cycle

we legit might have no time to adapt (not great)

7. practical shit we should start doing rn

invest in people early (schooling, retraining stipends)
way cheaper than mass unemployment later

share the productivity dividend (progressive taxes, ubi pilots, sovereign wealth funds) inequality causes riots, not automation itself

favour tech that complements humans (power-steering for your brain > straight-up replacing jobs)

rewrite social contracts asap (portable benefits, wage insurance, negative income tax) traditional payroll models gonna break fast af

8. expect political chaos (tech disruption always has political fallout, from chartists to occupy)
transparency + engagement > riot police

bottom line

history doesn’t repeat exactly, but it def rhymes

industrial revolutions show tech can mess things up short term but also massively raise living standards IF we reinvest and adapt fast enough

agi arriving within five years tho? that’s a speedrun we are NOT ready for

but at least history’s warning lights are flashing pretty damn loud rn 😂

what do you think? any good historical examples i missed? drop your takes 👇

[Alex Payne @AlexPayne]

> be me
> founder mode activated
> decide to work in coffee shop for “focus”
> new environment = productivity hack
> arrive at cafe
> spend 10 mins choosing seat with optimal lighting + plug access
> setup: MacBook, notebook, AirPods, aura of ambition
> no actual tasks yet
> order flat white with oat milk
> barista judges my order silently
> don’t care, I’m building an empire
> take coffee photo for Instagram
> caption: “grind mode”
> spend 15 mins choosing filter
> finally pick original
> open laptop
> stare at homescreen
> realise I left charger at home
> panic internally but keep cool externally
> AirPods in
> lo-fi beats on
> google “how to build MVP fast”
> read first 3 lines of article
> open new tab: YouTube
> watch 14-minute video on Notion hacks I will never use
> scroll Twitter for “market research”
> retweet AI meme
> laugh alone
> remember I need to pee
> don’t want to leave laptop unattended
> pack everything up
> go toilet
> come back
> someone took my spot
> sit next to toilet door
> cold breeze every 2 mins
> reopen laptop
> forgot what I was doing
> start fiddling with logo
> make it worse
> convince self it was a creative day
> close laptop
> feel satisfied
> no work done
> will try again tomorrow
> must need a better coffee shop

[Alex Payne @AlexPayne]

🚨 NEW: DeepSeek just open-sourced DeepSeek-Prover V2 (671 B MoE) on HuggingFace. No blog-post, just 163 shards / ~700 GB of weights 🤯

• 37 B params active per token
• 163 k-token context (YARN-scaled RoPE)
• Trained for Lean 4 theorem proving
• MIT-licensed, FP8/BF16 checkpoints

[Eliot Payne @eliot_payne]

#formalnight #bowtie #ovationoftheseas https://t.co/vWwG6LnxLo

[Eliot Payne @eliot_payne]

Please follow Ashleigh's dance page #coaching #dancer #dancing #teacher #dancers #dance #ballet… https://t.co/U3Mzlyeego

[Eliot Payne @eliot_payne]

#reunited #awesome good to see you tony_who77 !! #catchup @ Tenerife https://t.co/X3vmHphN6y

[Eliot Payne @eliot_payne]

#bettertogether #notlongnow #beautiful https://t.co/aFpQszoMAj

[Eliot Payne @eliot_payne]

#deckplan #bridge #map #ship #safetycommandcentre https://t.co/GOhdICg5Ru

[Eliot Payne @eliot_payne]

#sunny #Italy #shiplife #thelife #mylife photo credit @benmcloskey lol @ Manarola, Cinque Terre https://t.co/zZ3Elto0L4

[Eliot Payne @eliot_payne]

#flamingshots #everybodyshots #atomicbomb @ Baa Bar Nottingham https://t.co/UavwKOFS4I

[Eliot Payne @eliot_payne]

#ifly #skydivingatsea #2ndevergo #quantumoftheseas https://t.co/9XdwbcVKJ9

[Eliot Payne @eliot_payne]

#rollerskating #quantumoftheseas #scared @ Port Of Shanghai. China https://t.co/9I0E0lCH2V

[Eliot Payne @eliot_payne]

#perfectcapture #swing #beautiful https://t.co/Aq6QX0IPij