This guy gives a great breakdown of why the Madueke incident wasn’t a penalty.
He also has videos on why Saka’s hand balls weren’t hand balls.
The general issue with football is that the fans don’t understand the rules themselves.
nitter.cf/refereec21621?s=21
This is probably the most useful tweet I’ve seen regarding monetisation and why payouts have dropped so drastically for so many.
+$2,254 from X monetization for last 2 weeks
streak of 3 payouts when I am getting $2k+ from monetization
here's what I noticed changed with X monetization this time:
MOST IMPORTANT:
monetization for stolen/reposted content got mass-slashed
my friend got $180 for 12M impressions. all reposted TikToks and YouTube lectures. wild engagement inside every tweet. doesn't matter
X now detects it as stolen, low-quality content (especially if it's already been posted on X before)
meanwhile original creators with 2M impressions are getting 5-10x bigger payouts
the algo started heavily rewarding meaningful content with real involvement
stopped rewarding the views at all, gave a preference to bookmarks, replies, profile clicks
however, this original content is getting pushed hardly to TL which results in a low number of views
the best combo right now for maximum reach + monetization:
1: post your original tweet or article first
2: then find an interesting video or lecture from YouTube/TikTok
3: quote your own tweet with that video attached
the original post gets the monetization weight. the video QT drives the traffic back to it
here's which metrics ACTUALLY matter for virality according to the source code:
Final Score = Σ (weight × P(action))
Phoenix predicts 15 engagement signals for every tweet on your timeline:
heaviest positive weight: replies, profile clicks, DM shares, copy-link shares, follows
medium: reposts, quotes, clicks, dwell time
lightest: likes ← almost irrelevant
negative (SUBTRACT from score): not interested, block, mute, report
the exact numbers are hidden in the release. but what this means in practice:
tweet A: 10,000 likes, 12 replies, 0 shares
tweet B: 800 likes, 340 replies, 45 DM shares
tweet B massively outranks tweet A. likes without real engagement is noise
now add negative signals:
tweet C: 5,000 likes, 200 replies, but 30 blocks and 5 reports
those 35 negative actions can cancel out everything
tweet gets buried. account gets suppressed
the algo doesn't count views
it counts what people DO after seeing your tweet and mass-punishes you if that action is negative
especially I felt block/mute and report on myself
STORY: I released a tweet with personal observations on X algos when it got updated and open-sourced
it went viral. 2M views in 12 hours
but Nikita Bier said I was misleading the audience. a lot of toxic people on this platform took it too literally and sent a ton of negative feedback
I decided to delete it
few days later I caught a shadow ban. my impressions literally got dumped 5x overnight
with the highest probability from the toxics who mass-muted and blocked me after that tweet
one viral post with the wrong crowd cost me weeks of reach
WHAT ACTUALLY MATTERS:
stop chasing views. start creating content people save, share and reply to with real thought
the smartest play on X isn't going viral. it's going viral with the right audience
that's where the money is
BE YOURSELF.
1/ In 1849, thousands rushed to California chasing gold.
Most went broke.
The miners? Busted.
The guys selling shovels, jeans, and pickaxes?
They built generational wealth.
Levi Strauss didn’t pan for gold. He sold pants to the people who did and created one of the most iconic brands in history.
That’s the bottleneck principle.
And there are funds built around exactly this idea
8/
So, does this actually work in practice?
Here’s what the numbers show when applying a bottleneck approach (focusing on power/semiconductors + data centers/connectivity) vs the broad market:
The bottleneck categories have significantly outperformed the S&P 500 in recent years.
This is exactly why I believe the AI Bottleneck Fund thesis has strong potential, not by chasing hype, but by owning the critical infrastructure that powers the AI revolution.
The pattern is clear across 170+ years of innovation:
The biggest, most durable wealth has been created by those who supplied what everyone else needed - not just those chasing the shiny new thing.
I’m applying this exact principle today with the AI Bottleneck Fund thesis - focusing on power, chips, data centers, and connectivity.
I’ll be sharing the full journey in public: the research, the portfolio construction, the wins, and the mistakes.
If you’re an investor, founder, or operator in the AI space, I’d love to hear from you.
What bottleneck are you seeing right now that most people are missing?
Drop your thoughts below 👇
𝗙𝘂𝗻𝗱 𝗧𝗵𝗲𝘀𝗶𝘀: 𝗔𝗜 𝗕𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸 𝗙𝘂𝗻𝗱
𝐂𝐨𝐫𝐞 𝐈𝐧𝐬𝐢𝐠𝐡𝐭:
AI model capability is advancing faster than the physical and data infrastructure required to support it. The binding constraints, not the models themselves, will capture disproportionate value over the next decade.
𝐓𝐡𝐞 𝐠𝐨𝐚𝐥:
To validate whether concentrating in AI infrastructure, specifically power/semiconductors and data center/connectivity stocks would have meaningfully outperformed the S&P 500 over the period when AI capital expenditure began accelerating in earnest.
𝐓𝐡𝐞 𝐫𝐞𝐬𝐮𝐥𝐭𝐬:
A $10,000 investment in 2024 grew to $36,261 in power/semiconductors and $39,215 in data centre/connectivity by end of 2026 versus just $14,628 in the S&P 500. That’s roughly 2.5–2.7× the index return over just three years.
𝐓𝐡𝐞 𝐛𝐢𝐧𝐝𝐢𝐧𝐠 𝐜𝐨𝐧𝐬𝐭𝐫𝐚𝐢𝐧𝐭𝐬:
1.Power & Energy - AI data centers are projected to consume 8–10% of US electricity by 2030. Plays: utilities with data center exposure, grid equipment, nuclear
2.Semiconductors & Hardware - Beyond Nvidia: HBM memory, advanced packaging, networking silicon
3.Data Center Infrastructure - Cooling, REITs, construction & hyperscaler supply chain
4.Data & Connectivity - Fiber & low-latency networking, CDN & edge
𝐒𝐭𝐨𝐜𝐤 𝐔𝐧𝐢𝐯𝐞𝐫𝐬𝐞
Category 1: Power & Energy 25%
Category 2: Semiconductors & Hardware 30%
Category 3: Data Center Infrastructure 25%
Category 4: Data & Connectivity 20%
VST-6.25%
CEG-6.25%
GEV-5.00%
ETN-5.00%
PWR-2.50%
MU-7.50%
AVGO-7.50%
TSM-7.50%
ANET-4.50%
MRVL-3.00%
VRT-7.50%
EQIX-7.50%
DLR-6.25%
MOD-2.50%
AME-1.25%
LITE-6.00%
COHR-6.00%
NET-5.00%
CSCO-3.00%DATA & CONNECTIVITY 20% | ~$2.0M of $10M AUM
Moving petabytes of training data and serving inference at scale requires massive optical networking capacity. Fiber interconnect inside
DATA & CONNECTIVITY
LITE: high-speed interconnect for intra-cluster AI networking.
COHR: supplies coherent optics for both AI cluster and metro/long-haul applications.
NET: low-latency data delivery and AI inference distribution at the network edge.
CSCO: Ethernet switching and Silicon One ASICs for AI data centre fabric.
