Om
Former IBM, UBS, MUFG turned investor & tech founder.
Bullish on AI.
Tjenester
Artikler af Sam
-
What You Need To Know About The Energy Crisis
What You Need To Know About The Energy Crisis
The energy prices are not about to go down. What the heck is going on? How the energy market works A simple way to…
-
Most people calculate how much they save the WRONG WAY
Most people calculate how much they save the WRONG WAY
If by the end of the month you transfer £250 to your savings account, have you saved £250? If you answered “Yes”, keep…
2 Kommentarer -
What’s wrong with money saving robots?
What’s wrong with money saving robots?
The rapid boom of transaction round-ups and auto-saving apps promises an intelligent and effortless solution to save…
11 Kommentarer -
Where did my money go?
Where did my money go?
This is one of the most recurrent questions humanity has, yet it is still a pain to quickly find the answer. That's why…
4 Kommentarer
Aktivitet
4 t følgere
-
Sam Abrika delte detteI just tried Opus 5.5 for an hour and it's the GOAT. Smarter, cheaper, faster than all other LLMs. It can crunch and compare numbers across large reports without randomly misquoting half of them. And best of all: no more obnoxious Claudish load-bearing assumptions, Opus 5.5 finally writes proper English. Sonnet 5.5 and Haiku 5.5 are on the way. So when Dario said "we need to slow down AI development", he apparently meant "Anthropic will make mid-tier models so good you won't even need frontier AI".
-
Sam Abrika opslog detteWould you bet frontier LLMs can replicate how the best investors in the world actually think? Not a summary of their investment philosophy. I’m asking something harder: take decades of letters, interviews, investments, theses and public commentary, extract the convictions underneath them, then give it tomorrow morning’s news. Would you bet that AI can: - Decode daily new information through the same lens and arrive at the same conclusions - Grasp what this investor notices that everyone else ignores - Replicate when their conviction stregnthen/weaken I want to see how far LLM went, so I’m turning it into an open challenge. Who do you think is one of the smartest investors in the world? I’ll pick the most popular name and make the result publicly available next week: a live version decoding the news every day, free for anyone to use. One constraint: there needs to be enough public material about how they invest. Give me your top investor pick.
-
Sam Abrika delte detteI tried GPT-6 Astra and within a day it became my default model. It is genuinely impressive, the smartest model I've used, and the benchmarks agreed. Then it became a victim of its own success. Rapid mass adoption, usage tightened, performance degraded, and this Reddit Blender comparison is probably the best benchmark I've seen: Same model, same task, just one week apart. Funny... until you start using that LLM for critical workflows with money at stake. Frontier labs are gaming the benchmarks: launch model at full capacity → top the benchmarks → silently nerf it the following weeks. Imagine buying a stock based on its IPO prospectus and never checking the price again. Next unicorn idea: live LLM performance monitoring.
-
Sam Abrika opslog detteVCs put their portfolio on the website while an Equity investor would lie about a position at a dinner party. And the paranoia is rational because if your book is your thesis made visible, then a full position list is a confession. Except it isn't. Two investors owning the same company - One expects margins to expand - The other expects margins to stay flat, but bought because he thinks the valuation already priced the collapse Identical position, opposite drivers. Now hand that book to an AI and ask it to explain the strategy. It'll give you a pattern, sector trend, or maybe a factor decomposition that sounds impressive in a slide deck. What it won't give you is the second investor's quiet conviction that the market has already priced in a disaster that won't happen. Because that lives in his head, not the 10Q. The portfolio tells you what they own, the edge LLM cannot grasp is why?
-
Sam Abrika delte detteGPT-6 Astra is the new AI champion. Benchmarks look impressive, so before we declare AGI has arrived and retire to the beach while agents do the work, here is a simple test. Pick a stock, give it: - the last 2 years of filings - earnings calls - investor presentations - a live news feed Then ask: - What changed this week? - Which assumption just got weaker? - How is this company progressing relative to its competitors? - What would actually make me change my mind? Now run the exact same thing again. If it gives you roughly the same answer, for roughly the same reasons, well we're getting somewhere. Then (just for fun) try it on JPMorgan and its 400+ page annual report. If that all works, AGI is here and see you in Cape Verde 🏝️ .
-
Sam Abrika delte detteClaude Fable 5.1 is out. Anthropic used to increase token consumption by 25% with model upgrades. This time it's -25%, so good to see the curve finally going the other way. Can't wait to try it!
-
Sam Abrika opslog detteI don't think AI makes people think less, but rather changes how we think. Before AI, I would crunch problems in depth and sequentially to keep focus. Now I can have 6 different problems advancing in parallel. But that parallelism eventually collapses back onto one human brain. Every thread reaches a point where a human has to decide: what happens when the data format changes? would my underwriting silently fail? Is the model confidently solving the wrong problem? That's increasingly what my work feels like. Less time producing the analysis myself, more time jumping between partially solved problems to: - check whether my assumptions still hold - challenge AI-generated plans - zoom out and connect the dots across everything - prioritise between an endless list of things to analyse Once you've seen six things progressing in parallel, voluntarily slowing them down feels strangely like failure. At some point you either slow the machines down or drop scrutiny. I'm increasingly convinced that's the real cognitive risk of AI. Not that it removes our ability to think, but that its speed pressures us into deciding before we've thought enough.
-
Sam Abrika opslog detteI hear analysts & engineers say AI is making them dumber. Press the button, accept the output, copy-paste, next. My problem is almost the opposite. I can't keep up with all these AIs thinking faster than me. Because at any moment I have: - one agent building feature A - another debugging feature B - another testing why A won't speak to B - another specifying what comes next - another processing some SEC filing - another trying to automate a workflow I probably shouldn't be automating yet The problem is not just keeping the agents busy. It's keeping enough context in my head to make good decisions every time one of them comes back with a question, a choice, or something that looks correct but isn't. So I increasingly find myself as an underclocked CPU trying to sync 6 overclocked GPUs.
-
Sam Abrika opslog detteWhat I wish building software with AI was: Specify feature → build→ test it → iterate until it works I wanted to leave my pipeline running over the weekend, feed it the news and Twitter on my hottest stock, and come back Monday to see whether it analysed everything the same way I would. What building software actually is: Fix plumbing → fix a bug → discover 2 more bugs as a result → get swarmed with new issues and pull requests I didn’t want to deal with So that maybe, on Monday, I can finally continue working on the idea.
-
Sam Abrika reagerede på detteSam Abrika reagerede på dettegood news: solitude is down 70% YoY you're never really alone with claude
-
Sam Abrika reagerede på detteSam Abrika reagerede på detteYour last ChatGPT prompt was probably less than an hour ago. Your last post about billionaires wrecking the world was probably this week. I understand the week gave you reasons. The Artificial trailer landed, with Andrew Garfield playing Sam Altman walking into a bunker full of guns while telling you countries will fall... Then Jacob Coxon quit Anthropicic after 3 years of pretraining research there and at OpenAII, saying both companies are racing to self-improving superintelligence and gambling with your life. One of Anthropicc's own alignment leads put the odds of AI killing everyone within a decade above 10%. So you reposted the trailer. Added a line about greed at the top. Maybe asked ChatGPT to make it sound more human first. They've earned most of it. I'm sure the people running these companies deserve the beating, and you'll hear no defence from me. What I'm fed up with is the idea underneath it. That a handful of people at the top of the pyramid did this alone, and the rest of us just watched. You built their power. Every prompt you sent this year is revenue for the people you spent this week condemning. Day by day, night by night, hundreds of millions of us handed over the usage that put them where they are. I include myself. I co-founded an AI company. ETA Technologiesgies is part of the same wave, and I use these models every day. Before the next repost, take a step back and look harder in the mirror. Is it really true that a handful of people at the top are the only ones to blame? I asked myself that this week. My answer was no.
Erfaring
Uddannelse
Udgivelser
-
Liquidity Risk Implication of Intercompany Flows and Frictions
www.risk.net
Se udgivelseIntroduction of an emerging risk class to the banking industry.
Andre lignende profiler
Se flere indlæg
-
NEWNEX - Institutional VC Syndication & Co-Investment Platform
SFC Capital Secures €11.9 Million to Support 100 Early-Stage UK Startups https://lnkd.in/dbtubUMW Stephen Page Joseph Zipfel Angelika Burawska Ed Prior British Business Bank #SFCapital #UKStartups #StartupFunding #EarlyStage #VentureCapital #AngelInvestment #ScaleUp #Entrepreneurship #SeedFunding #Fintech #B2BTech #LifeSciences #ConsumerTech #InnovationEcosystem #TechForGood #GrowthTech #SEIS #EIS #LondonStartups #UKTech
-
David Craig
Systems don’t run companies—people do This was one of the most important lessons I share in "Bluebook", the story of Refinitiv and our transformation of a leading global financial data and analytics business. In Chapter 10, “People are your most valuable asset”, I reflect on managing through the COVID crisis. Like many organisations, we discovered our systems weren’t perfectly designed for the unpredictable, market data volumes and customer needs tripled over night. It was our people who stepped in and made the difference, often under difficult personal circumstances and fear of the unknown for themselves and their families. Looking after colleagues, creating an ownership mindset in the business, shouldn’t be reserved for times of crisis—it’s something we must do every day. Through any transformation, it’s your people, not just your plans, who make it happen. @Sam Brichalk summed this up well in her recent Raconteur article. http://bit.ly/3KvCxHl Fast forward to today: companies are making big assumptions about the power and impact of AI, especially Large Language Models (LLMs). LLMs are advanced artificial intelligence systems trained on vast amounts of text data, enabling them to understand, generate, and summarise human language with remarkable accuracy. While their potential is substantial, financial markets have already priced in $billions of transformational efficiencies and reductions in people and costs. This is, at best, premature—and at worst, misjudged. It shouldn’t take another crisis to remind us, or the next technology transformation: people make organisations run. Let’s not forget the lessons of the past. How is your organisation balancing the incredible but uncertain promise of AI, with the value of your culture and people? Jessica Spungin Rob Sartain Patrick M. Alex Fergus Shannon Shuemake Phil Wellard Craig Wymant Alison Grigg Eileen Lynch Neil Pabari Nathan Lynch
5 Kommentarer