Andy Wood (@andyhq) header
Andy Wood (@andyhq) profile picture

AI that works. News that's ranked. No filler. https://xitter.cf/t.co/VrgZrPJPUZ

UK
Joined July 2009
Photo by Andy Wood (@andyhq)Photo by Andy Wood (@andyhq)Photo by Andy Wood (@andyhq)Photo by Andy Wood (@andyhq)Photo by Andy Wood (@andyhq)Photo by Andy Wood (@andyhq)Photo by Andy Wood (@andyhq)Photo by Andy Wood (@andyhq)Photo by Andy Wood (@andyhq)Photo by Andy Wood (@andyhq)
Don't call it an AI winter yet. LSEG still expects 31.6% year-on-year S&P 500 earnings growth this quarter. The sell-off is sorting winners from spenders. Which suppliers survive the shake-out?.... Nasdaq has its worst day since June as AI spending fears hit chip stocks The Nasdaq fell 2.86% to 25,437.00 after weak results from Intel and Amazon, with Intel down 28.6%, AMD down 10.9% and the chips index down 7.2%. Investors are now openly asking whether AI capex will pay back, and how quickly. finance.yahoo.com/markets/st…
3
Funny how the timeline for superintelligence and the timeline for a funding round never line up. Which one slips next?... OpenAI cuts its funding round target by $10bn to about $20bn The FT reported that OpenAI trimmed its raise from $30bn to about $20bn and its target valuation from $1.4tn to $1.25tn, with the first close expected below $20bn after some investors cut allocations. It lands as Anthropic heads for its own listing and AI stocks slide. investing.com/news/economy-n…
2
1
17
S&P said what too few analysts will. It's unclear when Meta's AI revenue will beat its rising costs. Capex guidance of $135bn to $145bn deserves exactly that scrutiny. Will equity analysts follow the bond people for once?... S&P cuts Meta's credit outlook to negative over AI spending S&P Global Ratings kept Meta at AA- but moved the outlook from stable to negative, citing elevated capex and negative free cash flow. It also questioned how fast Meta can monetise AI. Meta guides to $135bn to $145bn of capex this year and plans to raise $95bn. cnbc.com/2026/10/30/meta-out…
14
A research report that takes longer to check than the job took by hand has saved you nothing. Here's how to test the research modes on a job you've already done, with a red-card rule for the fails. aiprompthackers.com/p/how-to…
22
Before Monday, pick the AI tool your firm would struggle without for a week and write four lines: What it does. Who owns it. Which model sits underneath. When someone last checked its output. If the third line is blank, the brief has the supplier email for that. readanswerable.com/p/ai-regu…
1
1
17
Weekend job: delete the five AI chats you'll never open again. Rename the one you will.
80
The Digital Hum...
1
70
Don't assume the research report with 80 sources won. In the PwC preprint, factual accuracy fell by about 42% on average as agents were allowed more tool calls. The links kept working the whole time. aiprompthackers.com/p/how-to…
2
2
63
OpenAI's new Model Spec says GPT should act on your "intent" when your instructions run out. So a vague prompt doesn't fail any more. It guesses. End prompts with: "List the assumptions you made."
3
2
52
Before you ask AI to fix a document, tell it who's going to read it. What's happening: edits come back generic because the model can't tell whether the reader is a client, a board or a tired colleague. Three things to add: 1. Who the reader is and what they already know. 2. What you want them to do once they've read it. 3. A request for the three sentences that reader is most likely to question.
1
37
Good intent, wrong level. State-by-state frontier AI rules mean 50 versions of the same audit. This needs federal law. New York may force the issue, which is maybe the point... New York senator proposes RAISE Act 2.0 with compute limits and licensed AI auditors State Senator Andrew Gounardes's bill would cover any firm spending more than $1bn on AI development over three years, require a state-licensed "frontier examiner" and let a new state office limit training compute. Penalties would reach $1m a day for a first violation and $3m a day for later ones. nysenate.gov/newsroom/press-…
27
Google Cloud grew 103%. The stock still had its worst day ever. Free cash flow went to −$9.07bn on $51.4bn of capex in a single quarter. The market has stopped grading AI on growth and started grading it on cash. Not advice, just the maths.
63
Publishing the spec is good practice and other labs should match it. Values give regulators something to test against. The next step is independent checks that the models follow it... OpenAI rewrites its Model Spec around five values The public rulebook for how OpenAI's models should behave is now organised around faithfulness to human intent, harm minimisation, honesty and transparency, human agency and care. It adds guidance on context, oversight and agency and retires much of the old style guidance, which changes how ChatGPT should handle grey-area requests. openai.com/index/updated-mod…
1
23
Meta scaled back plans to swap staff for AI agents. Why? Its people spent too long fixing the agents' mistakes. If Meta can't absorb that cost, your 15-person firm can't either. Time every fix for two weeks before you buy another seat.
1
1
142
So Reddit's plan to fight AI killing search traffic is to make itself harder to visit. That's like a high street fighting online shopping by bricking up the doors. Bold. Stupid, but bold... Reddit shifts to a logged-in experience as AI search cuts its traffic Reddit said search referrals fell by more than a million a week in October and warned daily users could drop by a low double-digit percentage if the trend continues. It'll test login requirements in some non-US markets this quarter, a stark example of AI answers starving the open web. axios.com/2026/10/29/reddit-…
1
72
Google lost half a trillion dollars in a day and still grew cloud revenue 103%. Only in tech can you double a business and get treated like you've set fire to a care home. Investors, eh... Alphabet has its worst day ever, losing about $500bn Shares fell nearly 13%, beating the 11.63% record drop of March 2020, after Alphabet reported free cash flow of minus $9.07bn and search ad growth of 12% that missed forecasts. It's the clearest sign yet that investors will punish AI spending that outruns cash generation. cnbc.com/2026/10/29/alphabet…
1
37
AWS beat forecasts at $45.98bn while revenue overall missed. Demand for cloud compute is real. The open question is whether $77.3bn of quarterly capex earns a return... Amazon misses on Q3 as AI spending drives a $42.4bn cash outflow Revenue of $201.94bn and earnings of $1.89 a share both missed forecasts, while AWS grew 39% to $45.98bn. Quarterly capex hit $77.3bn against $35.1bn a year earlier, trailing 12-month free cash flow was minus $42.4bn and shares fell more than 6% after hours. ir.aboutamazon.com/news-rele…
2
1
55