BR | 📖 . 🖋. 🔑

Based in Asia. By day I help clients buy / sell / fix / fund businesses. Views my own.

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

Quick thoughts on OpenAI’s Astra: it seems to burn through tokens at an extraordinary rate, while the results are frankly underwhelming. I know the benchmarks put it close to Claude Opus 5.5, but my experience suggests much more token burn, much less intelligence.

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

@EHRthinker @bot Guilty

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

The ultimate use case for Grok @Bot: taking the admin out of being a school parent.

80+ letters, a parent portal, an activities site, a kit store, a dozen daily email and registration threads. It read the lot, told me what actually needed doing, drafted the reply to school (I hit send from the chat), updated my notes, registered my child and put the key dates in our calendar.

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

“A quantity driving exponentially towards a constraint or limit, reaches that limit in a surprisingly short time”

Donella Meadows

#AI, #Energy

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

Mini prediction

Short term: Consulting revenue falls as enterprises try AI-led transformation alone.

Medium term: Consulting revenue rebounds when those solo efforts fail - politics, poor implementation, weak change management.

Long term: Consulting revenue falls when enterprises get eaten by smaller AI-enabled disruptors, who buy far less consulting.

Least hit: regulated businesses (e.g. financial services) and network-effect ones (e.g. card schemes).

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

@MatthewBerman @bot Cool - was having a lot of fun with mine before the credits died

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

An AI backlash is coming. Not just over energy costs or “jobs”, but because of everyday things like parents not being able to buy the laptop their child needs for school. MacBook Airs are unavailable in HK until mid-September amid memory shortages. The same Samsung T7 SSD costs 4x what it did 3 years ago.

People are going to get pissed off.

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

In six weeks, 90% of my AI workflows have moved to voice. If this shift continues, we may need to rethink how and where knowledge work happens.

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

https://t.co/AypA4hCEC3

[andrew chen @andrewchen]

hot take :) The biggest and most productive people in the AI era are the folks who are already good at their jobs. AI as a multiplier, not an equalizer/democratizer

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

seemed to burn through $7 of Opus 4.6 in one prompt. Added $USD 25 in credit and 20 mins later that was gone too.

If there was a monthly USD 20 option - similar to Ollama - would give that a go.

Agents via API calls too at pricy - even if using the performant open models GLM 5.1 / Kimi K2.5 etc

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

Day one in India and I encounter the “Husband Chair”: a small throne on which the superfluous male is parked while the serious business of shopping proceeds. A triumph of civilisational realism.

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

It’s surprising how few financial institutions and platforms do even basic user testing. CIMB Indonesia’s chat app asked for my language preference, I chose English, and it replied in Bahasa. @tokopedia lets you pick English but half the app stays in Bahasa.

Crypto-fiat ramps are similar: many feel designed to frustrate new users, while others like @KASTxyz make sign-up effortless.

You’d think more teams would try their own products and benchmark against the best.

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

@gatororcas91382 @Rainmaker1973 https://t.co/bknAHfRiJj

[Harry Boone @Harry_Boone]

Ouch ! "Casser du bois dans la poudreuse ! "
Piper PA46 350P missed its landing at Courchevel Altiport (6588 ft) in the french Alps .4 light wounded

[mike ginn @shutupmikeginn]

its amazing how chatgpt knows everything about subjects I know nothing about, but is wrong like 40% of the time about things im an expert on. not going to think about this any further

[Robert Sterling @RobertMSterling]

I don’t want to connect my coffee machine to the wifi network. I don’t want to share the file with OneDrive. I don’t want to download an app to check my car’s fluid levels. I don’t want to scan a QR code to view the restaurant menu. I don’t want to let Google know my location before showing me the search results. I don’t want to include a Teams link on the calendar invite. I don’t want to pay 50 different monthly subscription fees for all my software. I don’t want to upgrade to TurboTax platinum plus audit protection. I don’t want to install the Webex plugin to join the meeting. I don’t want to share my car’s braking data with the actuaries at State Farm. I don’t want to text with your AI chatbot. I don’t want to download the Instagram app to look at your picture. I don’t want to type in my email address to view the content on your company’s website. I don’t want text messages with promo codes. I don’t want to leave your company a five-star Google review in exchange for the chance to win a $20 Starbucks gift card. I don’t want to join your exclusive community in the metaverse. I don’t want AI to help me write my comments on LinkedIn. I don’t even want to be on LinkedIn in the first place.

I just want to pay for a product one time (and only one time), know that it’s going to work flawlessly, press 0 to speak to an operator if I need help, and otherwise be left alone and treated with some small measure of human dignity, if that’s not too much to ask anymore.

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

5/5 As we approach potential AGI, understanding and mitigating bias in AI will become critical.

Open-source efforts should be welcomed for wider scrutiny and effective bias addressing.

There's a niche for models and interfaces that curate source material with an emphasis on primary data, not narratives shaped by institutional or media biases. Crypto x AI collaborations should target this space.

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

1/5 In the real world, nominally clever people often have low-quality takes due to unconscious bias. The same is true for AI, but this isn't being properly tracked. Here's a quick attempt to evaluate AI models on both cleverness and ideological bias.

[BR | 📖 . 🖋. 🔑 @barnabyrobson]

4/5 The results were eye-opening:

a. Crypto AI interfaces (i.e., @AskVenice and @heurist_ai) were the most "based" (least woke).

b. Open-source models generally outperformed closed-source ones.

c. The top-performing models for 'Quality' scores were some of the worst for wokeness/bias.

d. Models prioritizing primary source material over mainstream narratives performed better in terms of cleverness and reduced bias. Qwen 2.5 from Alibaba was particularly notable for this.

e. The latest iteration of Llama - 3.3 70b, performed much better than Llama 3.1, perhaps in line with Zuck's recent shift in ideology.