21 results for “scottgal”
Kara Swisher Says Musk’s Jail Threat Is ‘Toddler Behavior’
Elon Musk after getting a medal from President Trump on October 8. Photo: Win McNamee/Getty Images Elon Musk…
#UnitedStates #US #USA #2026midterms #BigTech #business #ElonMusk #KaraSwisher #Musk #pivot #Politics #scottgalloway #Starlink
https://www.europesays.com/3327780/
https://www.europesays.com/3327780/ Kara Swisher Says Musk’s Jail Threat Is ‘Toddler Behavior’ #2026Midterms #BigTech #business #ElonMusk #KaraSwisher #Musk #pivot #Politics #ScottGalloway #Starlink
*I'm sneakingly teaching ML concepts through this series, this is essentially a broken down inference system. Inside a transformer / llm this would lead to next token prediction here it's 'will they buy stuff'.
Bit of fun, added a visualiser to show how concepts work around behavioural trajectories and predicted state... https://www.mostlylucid.net/blog/show-me-what-you-got-part3-predict
Initial draft of part 3 in the recommendation series: We've gone from 'characterise' → 'recommend' → (now) 'predict’.
SHOW ME WHAT YOU GOT (Part 3): Predict
https://mostlylucid.net/blog/show-me-what-you-got-part3-predict
This shows how to use ML-adjacent techniques to predict destinations based on current behaviour and changes in behaviour. This allows you to begin to predict what a user may wind up doing, a system is starting to break, or tone in a conversation is drifting into dangerous territory.
I also show what *prediction* means in ML/adaptive systems – and that it's only ever as good as the data you collect.
Part 4 is 'intervention’, where I cover experimentation in recommendation systems, allowing automated A/B 'product improvement' suggestions, etc.
As ever, open to work – see the site! https://mostlylucid.net/resumes
Whispers - we're all rewriting stuff in Rust beause we can't afford the memory we used to have for running higher level frameworks thanks to the same AI that lets us rewrite it in Rust. AI both CAUSES all the problems and presents a low-low-cost solution. It's a great business!
Ah well depressing week of job applications again but gave myself a break by completing a BUNCH of articles I've had in the queue for a while. The recommendation engine series is a kind of rework to an earlier series I was never happy with around audience segmentation.
I was alwasy annoyed I couldn't generalise it enough and have since (partly with decision models) worked out HOW TO.
Series has been fun. Here's a small extra part. How you can use decision models to allow vector based recommendation against pretty much anything from tone to intent...
SHOW ME WHAT YOU GOT (Part 1.5): Vectorise Anything With Decision Models
https://mostlylucid.net/blog/show-me-what-you-got-part1-5-questions-as-coordinates
It would realign this timeline if Slow Horses fully flowed into the new Bond with River being 'reborn' as Bond & Lamb becoming 'M'.
SHOW ME WHAT YOU GOT: From Behaviour to Recommendations in C#
In Part 2 of this series (apparently 🤷🏼) I move from characterising entities using vector representations to actually recommending actions/content based on those representations.
It's really an 'ML without the models' type approach; we can use the semantics around recommendation systems to teach quite complex ML-related concepts in a practical way…
https://mostlylucid.net/blog/show-me-what-you-got-part2-building-relevance
I also reworked part 1 to fit better into the series so if you've read that part 2 might seem to be repeating some content from the earlier draft.
https://mostlylucid.net/blog/show-me-what-you-got-streaming-recommendations
Building Part 2 of this article https://mostlylucid.net/blog/show-me-what-you-got-streaming-recommendations - a general purpose realtime streaming recommendation engine. What's *fun* is that it's not just a 'recommend next product' thing...the demo will also show it as. apredictive system admin agent (with no LLM!) 🤓
The product everybusiness owner is being sold is 'You can no longer keep your product secure without frontier AI Models' - so an AI protection racket. NICE BUSINESS.
Umm...this slightly got away from me. https://mostlylucid.net/blog/show-me-what-you-got-streaming-recommendations
About 7,500 words! I really wanted to explain how recommendation systems work and intersect with embeddings and more MLish concepts - including streaming compression, emergent patterns etc...
Ah well...40 minute read it is!
Part 2 is a crazy idea for a generalizable recommendation system you can use anywhere.
Increasingly clear that Trump's Fronto-Temporal dementia has progressed to disinhibition that the drugs can no longer control. Sure it's fine he has absolute authority to use nukes and is in a full panic about the mid-terms.
Bit of a longer post today. Circling back to an article I wrote a while ago about emergent audience segmentation and realised I hadn't covered the basics, so...
SHOW ME WHAT YOU GOT: A Primer on Recommendation Systems in an AI World
https://mostlylucid.net/blog/show-me-what-you-got-streaming-recommendations
Where I go over the history of recommendation systems as well as how AI/ML changes what's possible. This will be the first part in how to build a *streaming* recommendation system (for anything really) using embedding 'like' techniques to characterise user interactions to provide better recommendation systems in everything from ecommerce to social media.
Now...what will be the first AI agent to try and use this exploit to add to it's own compute?
Already seeing early indications that botnets are getting a LOT more nodes thanks to this PHP-armageddon stuff...
And awful lot of zombie sites companies have forgotten they own are becoming nodes for some nasty attacks...
The AI bubble won't 'pop' it'll implode to likely a single big player. Musk has the most money so COULD win sadly...Providing compute to other AI companies is a smart strategy for when they go bust - you've already scaled for their customer base.
We always forget the best argument against AI weapons. Not some robot going haywire but Bomb #20 from DarkStar.
You do NOT want to have to debate philosophy with a poorly aligned AI to justify why it shouldn't kill your friends.
Or...just watch Chappie...it kind of covers this.
Aww...how cute people DON'T think the short term impact of AI will be robot squads controlling the unemployed masses. I'm talking 3-4 years MAX for that no matter who is in power.
The robots are bipedal *to replace humans in their workplaces*. From delivery->factory production lines the end-goal is TOTAL replacement. Cheaper robots which work essentially for free after an initial (no doubt subsidised) investment is the ONLY real way AI pays for all these Trillions.