Peter Fuller

Introducing Tracelight 1.0 After six months, we are live with 5 of the 10 biggest management consultancies in the world, as well as PE funds and asset managers with over $600B in AUM. Tracelight is the best AI add-in to Excel. In testing against the leading alternative add-ins, Tracelight 1.0 was: - The most accurate - 2x faster building analysis on average - 20x faster at finding human errors in complex spreadsheets We’ve been building with early adopters and enterprise partners for months. You can now reach a completely new level of performance in Excel. That’s why today we’re proud to launch a major new version. To celebrate, we're giving 1 month of Tracelight Pro for free to anyone who comments LAUNCH below or TAGs a friend who needs this

@mikeknoop That's an odd final sentence. Humans absolutely depend on their harness. You wrote this post on a computer, and shared it over the internet. Those tools are not 'internal' to you, but without them you would solve far fewer problems.

LLM systems swallow harness progress. The most general/universal LLM innovations migrate from client-side harnesses to server-side tools. Innovation typically happens first inside the harness. For example, AI reasoning was originally a harness around GPT-3 ("let's think step by step"). This approach worked so well that it migrated behind the API as a tool (competitive reasons were also a factor; but general utility dominated). Many wouldn't think of AI reasoning as a tool but it definitely is (it's a tool to do natural language program synthesis -- but that's another topic). The same happened with code interpreter which started out as a client-side harness and moved server-side. These tools are made available at inference time to the model alongside specific training to teach the model when and how to use each tool. Because of this, the line between tool and model can get quite blurry. Best to consider such tools as "internal" to the LLM system. This is actually a good test of how general a harness feature is. If a feature remains "stuck" client-side, say inside codex or claude code, then it's likely very task- or domain- specific. Client-side harnesses typically encode a lot of human G factor for specific domains. Whereas tools, due to usage pressure of frontier LLMs, are required to be as general as possible else they wouldn't make the cut. So if you care about measuring AGI it's a good idea to pay attention to default LLM system capabilities behind high usage LLM APIs. And if you care about bleeding edge research ideas, such as RLMs, it's a good idea to pay attention to harness innovation. Ultimately, AGI will not depend on a harness in the same sense humans don't depend on a harness.

Tbf every problem he diagnosed is completely correct, I just don't think the path forward is exactly what he thinks it is

In the seven days since @andrewchen predicted that spreadsheets are over because of no versioning or connection to code, we have released: - spreadsheet versioning - connections to code (web dashboards built off of spreadsheets) Life comes at you fast

@amaankahmad Love it. Congrats!

We're excited to announce Pathfinder's $4M seed funding led by Gradient and Reach Capital to put education back into the hands of parents  - pathfindercard.com/seed

@thejamesmcaulay @TracelightAI To be fair to him there's a lot of releases to keep track of

@peterfuller23 @TracelightAI This is so sick. Jeremi kept that quiet in our call last night!

From today, @TracelightAI turns your spreadsheets into beautiful, shareable webpages. I’m so excited about this feature, it’s magic. You could already create Powerpoints or Word docs with Tracelight. But now you can build interactive webpages that tell the story of your analysis so much more powerfully. It’s not just about saving time - although it definitely does that. It’s about differentiating yourself. Once your clients and team see dynamic artifacts like this, they won’t want to go back to the old way of doing things. Over the next few days, we will be making Tracelight’s webpages even better - adding chat, and native citation to the underlying Excel. Welcome to the future everyone 🙂

You can check it out here for free: tracelight.link/-compare

Someone changed a number in your model. Good luck finding it. Today we're launching something to fix that - Spreadsheet Compare. Drop two versions of the same spreadsheet into Tracelight's platform, and you get: - A view of all the changes between the sheets: what cell values and formulas were before, what they are now, and any structural changes (new rows or columns) - An AI summary of the differences This solves so many problems for our users. To name just a few: - Checking what changes your associate made to the model - Figuring out the drivers of valuation changes between periods - Making sure no one has made unauthorised changes The more powerful AI gets, the more you’re going to need this. Model comparison and versioning is one of the key pieces of infrastructure that will allow us to hand off more modelling work to AI. You can try it now on @TracelightAI's platform - link in the comments!

@s8mb I don't think William Hague's point is valid. Barack Obama spent his early career organising at a local level - being a 'local caseworker'. Didn't stop him.

We should give MPs large caseworker teams (like 25 people each) who can be professional social workers for them and stop the MP themselves having to do it. It’s one of the only things I can imagine that would be popular *and* make it less bad to be an MP.

@andrewchen If someone's homework reply is 15 pages of meandering AI-generated slop, you just shouldn't hire them It's still a good filter

In recent years interviews for execs, product managers, marketers, etc have started to add a homework step so that people can show work output. It can be super helpful signal In recent weeks these homework replies have become overwhelmed with AI slop. Instead of 2-3 succinct pages reflecting a career of deep thinking, you get back fifteen pages of meandering ideas that anyone could generate What's the best way to address this? The best way, of course, is to actually work with them on a work trial so that you can really get a feel for how they act; however this is difficult because it takes a lot of time for them and for you so you have to reserve it for the end of your hiring funnel A more scalable idea is to ask people to present their homework in a recorded presentation so that they actually sign off on everything that they say and it's possible to ask them interactively later If you go much more structured, you risk offending the highest-end talent Other ideas? Open to the thoughts

@taimurabdaal Doing god's work here taimur

We’re excited to announce that @join_ef has raised $200m of fresh capital, including $130m into our management company at a unicorn valuation, to be the natural home of the world’s most ambitious people in the Age of Entrepreneurship 🧵

@nathanbenaich AI will spread to every corner of the economy but still British Airways will stand firm When the BA mobile app and website have been refactored we will know that the last bulwark has been breached

ba it webmaster is not a sucker for details

@s8mb Nothing standing in your way! Please share when done

I’m quite excited about the prospect of replicating HelloFresh with an AI + Ocado + knowledge of my diary to calibrate how much cooking I’ll be keen to do on a given evening.

Always been sceptical of the claim that 'with AI, the juniors aren't going to learn the mechanical basics of the job, so won't become good seniors' Seems easy to build AI that teaches? And would be better than current training. Cursor could have 'learning mode' etc

@VictorTaelin Good luck!

I think for the first time in all these years I feel like I'm finally about to release something that is ready to, and *will*, be massively adopted... ofc it could just flop (and that's fine too!! expected, even), but I never felt like that and that's a very cool feeling

'Winter Olympics' possibly the greatest branding of all time This thing is not serious, these people are frolicking in the snow

@NathanpmYoung I believe AISI has about 300 employees. You want 3000?

The British Government could 10x its AI Safety Department. We are legitimately world leading.

@coop4real follow me pls so I can send access!

@peterfuller23 LAUNCH

Introducing Tracelight 1.0 After six months, we are live with 5 of the 10 biggest management consultancies in the world, as well as PE funds and asset managers with over $600B in AUM. Tracelight is the best AI add-in to Excel. In testing against the leading alternative add-ins, Tracelight 1.0 was: - The most accurate - 2x faster building analysis on average - 20x faster at finding human errors in complex spreadsheets We’ve been building with early adopters and enterprise partners for months. You can now reach a completely new level of performance in Excel. That’s why today we’re proud to launch a major new version. To celebrate, we're giving 1 month of Tracelight Pro for free to anyone who comments LAUNCH below or TAGs a friend who needs this

@shivmalik @Markos_mom Shiv I'm a supporter but these mostly seem like good questions you should answer. And you shouldn't complain about someone being an anon on twitter imo

@Markos_mom Sure buddy. You can create straw mans all you want but it’s a game I ain’t gonna play with an anon. Have a nice weekend.

Two entrepreneurs want to build a million-person city on Suffolk farmland. - Permanently affordable homes. - "Permission, not subsidy." (their words) The numbers tell a different story entirely IMO. I wrote the bids for half the Homes England feasibility studies. I've delivered housing and infrastructure at scale. Happy to argue any of this. The model proposes 400,000 homes via Community Land Trust, meaning resale prices are capped forever. That's 235x more CLT homes than the sector has ever delivered, ALTOGETHER. When asked how, the founder's answer was one word: "Community land trust." No mechanism. No understanding. Just dismissive platitudes. Cost estimate for 400,000 homes plus hospitals, schools, trams, rail, utilities, nuclear reactors: £130-200 billion. That's conservative. Assuming we were a nation who were actually able to deliver infrastructure properly. They're raising £200 million. For planning. That's 0.13% of what's needed. Investors at £100bn+ scale need 5-15% annual returns. 5% on £100bn = £5 billion per year. Where does £5bn annually come from if prices are capped forever? It can't be government, because they said they weren't asking for subsidy. So it's private investors. The saudis? Quataris? Kuwait sovereign funds? You cannot promise investors market returns AND promise residents prices won't rise. The maths doesn't work. When pressed, the founder admitted: "The SEZ squares the circle." Special Economic Zone. That means tax exemptions, business rates relief and regulatory carve-outs. That's not "no subsidy." That's billions in foregone tax revenue. They want to break ground within this Parliament. ~4 years. That probably sounds really ambitious and achievable if you've never actually been involved in infrastructure planning and delivery at scale. It will take 4 years to commission and deliver a realistic feasibility study at that scale. Never mind the collosal undertaking of putting that amount of land through compulsory purchase orders. The entire UK builds c.200,000 homes per year total. They're proposing double that in one location. Where do the workers come from? The materials? The supply chains? And that's just for the housing, never mind the collosal infrastructure required. Let's look at the land. What do they control? Nothing. By their own admission, 50% of the land is owned by three lords, an Arabian sheikh, and two solar farms. No options. No agreements. A map with a circle. Their plan is to lobby for Development Corporation powers, and use compulsory purchase powers. That's not "permission not subsidy." That's asking the state to forcibly acquire hundreds of millions in private land on their behalf. The founders have stated their company would receive 80 acres of city centre land as "reward." Agricultural land trades around £10k/acre, vs prime city centre in a million-person SEZ at £2-5m/acre. That gift could be worth £150-400 million, and under SEZ rules, there'd be reduced taxation on profits. I believe the founder stated the trust wouldn't be controlled by residents because "democratic choice often runs counter to long-term interests." That's the opposite of how CLTs work. So what do you have? - Fantasy maths - Fantasy land - Fantasy permission - Millions in profit for the two entrepreneurs and their backers - Zero growth for residents. Locked in to low value housing forever - A shift in the overton window on new towns - The two entrepreneurs and their advisory board positioned as the go-to consultants for new towns and cities (the likely grift IMO)