Mads Primgaard

@Domestique___ Riding defensively on a stage where he’s out numbered and wearing the leaders jersey. You fucking imbecil

'Jonas rode very defensively all week' - 🇧🇪 Remco Evenepoel rues crash and missed chances in 🇪🇸 Catalunya 📰 domestiquecycling.com/en/news/jonas-… 📸 Cor Vos

@iammarkmonroe I wish you’d just talk markets and tech once a month. I’d say what I’ve learned from you is hard to distil into a video or a specific teaching. Value for me is in the loads of spontaneous nuggets when you’re just speaking your mind. Do it!

If I was to come back to Youtube, what would you everyone want to learn from me that I haven't already taught?

@faustocoppi60 Lol just like when Pogacar crashes, right? Stop with the excuses.

Remco Evenepoel finishes 1:37 behind Vingegaard and 37 seconds behind Lipowitz. For sure not at 100% after the crash but this still can't be good for his morale. #VoltaCatalunya105

@tomdanielson I will never cheer immaturity. Stop normalising shitty behaviour. He’s emotionally immature like few other 😂

Look at this headline! This headline reads like a TMZ one about Justin Timberlake like Remco was drunk in the finale of today’s stage. Nothing “bizarre” about a guy so strong he drops everyone in a crosswind and tries to win despite the guy who has beaten him at all the last grand tours sitting on him and not helping. Also nothing wrong with showing emotion in the moment. Tadej was on Netflix in this same scenario with Remco telling Jonas to f’ off for sitting on. Bike racing is hard and human beings are emotional when they give everything.

@iammarkmonroe I’ll take the calls when they fill the gap, anyway.

Uh Oh!! youtube.com/watch?v=uJcf2U…

@nflshowet @nflming @Thomasqvortrup Det er ikke i orden det der.

Alt godt har en ende 😢 Det gælder også for ⁦@nflming⁩ og ⁦@Thomasqvortrup⁩, der er nået frem til at skulle sige farvel og tak … til 2025-sæsonen 🥳 Vi ser også frem mod combine, free agency og draft. Og så er vi tilbage til marts 🏈🍿 #NFLdk podcasts.apple.com/dk/podcast/nfl…

@AndreasSteno @berlingske Jeg er meget enig ^

@berlingske Videoen er både faktuelt forkert (husk at fratræk alle de importerede chips, når du regner bnp bidrag fra AI) og så ovenikøbet så rygende biased at det ikke er til at holde ud.

Donald Trump har satset på kunstig intelligens og kryptovalutaer som vejen til USAs verdensherredømme. De seneste ugers massive kursfald betyder, at hele projektet vakler. Læs mere her: berlingske.dk/oekonomi/trump… Video: August Plet Bang Redigering: Sofie Rose Hedelund

@nikoliasgoninus @NFTLunatic Unfortunately the platform is very political..

@NFTLunatic I’m still testing the video calls. I prefer Falstaff personally.

I upgraded to $DUOL Max. Have some great feedback for the Duolingo team soon. Long way to go on their AI tutors but they are super promising.

@nikoliasgoninus What’s your position size in duo? Here mine:

Bought more $DUOL and $PATH. Love cheap stocks.

@FetzInvests @TrendSpider Did you put in 2 hours everyday for months on Duolingo Max? And you didn’t learn? Say you’re an imbecile without saying it.

@TrendSpider The sad thing is you didn’t even learn ¡Adiós! from $DUOL

¡Adiós! 👋 $DUOL

@QualityInvest5 @CCM_Brett Cheap is not a reason to buy tell us what they will be in 20 years instead

@CCM_Brett you really should dude, stock is SO damn cheap

Why should I not buy $ADBE at $300

@QualityInvest5 They don’t train on customers data unless they bought it.

No you're just flat out wrong man 1) they have a shit load of data, they have 40M customers some of whom have been customers for decades ... that is BOAT LOADS of data 2) they have 40% FCF margins and 37% EBIT margins. That's higher than many big tech names including Google 3) they have 60% market share of the creative software industry and 90%+ of the world's professionals use their software "read between the lines" wtf are you on about?? $ADBE

You're off on Tesla. Their moat isn't self-driving software. There will be many more of those eventually. The moat is the production platform. Whoever produces faster, better, and cheaper will ultimately have the advantage in scale and price in autonomy. The next-gen robotaxi platform will likely be even more insane.

Do you think $ADBE has a wide moat? A clip from my recent video (link in bio)

@kazzazapple @PhilippEffkema1 @ondrejslunecko Biggest problem for casual learners is motivation. Biggest market is casual learners. Gamification is a very important part of this.

Fair point “fluency” was an overstatement on my part. After years of learning languages and using Duolingo since 2014, it isn’t enough even for basics: you struggle to reach real A1 in speaking, listening, or writing, with maybe a bit of A1 reading and vocab at best. Any decent formal course gets you much further. You really think Duolingo with its current offering can compete with formal study at a university where they actually teach languages professionally for real-world use? Come on. They need to take the language-learning part as seriously as the gamification, or competition will eat that advantage.

1/ What if I tell you that Duoling $DUOL is going to trade north of $636/share by 2030? And these numbers might be conservative! Let's dive into HOW this is going to happen 👇

@papasofrito @TheCryptoLark @RemindMe_OfThis in 5 years

@TheCryptoLark With all the advances in technology there is better places to invest.

We all know Duolingo as the app we use to learn Chinese to gain an edge in the trenches But under the hood, it's an $8B company with 50M daily users and AI-driven learning that's scaling fast The growth curve is real, and the monetization is accelerating Is $DUOL a good investment? ⤵️ thewealthmastery.io/duolingo-inc-n…

@aatambtc @TheCryptoLark Watch this change over the coming years. Even now, I disagree. I learned Spanish fluently, then didn’t speak it for 12 years, and now I’m picking it back up so fast. What helps me the most? AI conversation. We’re just getting started. You’ll be wrong here and miss it.

@TheCryptoLark Duolingo is somewhat good for someone who have zero knowledge about a particular language. It’s the worst if you’re somewhat familiar with a language.

$AMD MI450| $NVDA Rubin| $GOOGL TPU Comp🧵 The @AMD Instinct MI450 emphasizes open ecosystems, memory density, and power efficiency via chiplet design and TSMC's 2nm node w/ lowest TCO. NVIDIA's Rubin (successor to Blackwell) prioritizes raw compute and bandwidth in a proprietary stack(CUDA). Google's Ironwood TPU (7th-gen ASIC) focuses on TensorFlow-optimized inference at pod scale These 3 will be released in 2026 Architecture AMD: CDNA 5 (UDNA-based chiplets) NVDA: Rubin (monolithic/early chiplet) Google: Custom ASIC Process Node AMD: TSMC 2nm NVDA: TSMC 3nm Google: TSMC 3nm Memory (per Chip) AMD: 432 GB HBM4 NVDA: 288 GB HBM4 Google: 192 GB HBM3e Memory Bandwidth(per Chip) AMD: 19.6 TB/s NVDA: 20 TB/s Google: 7.4 TB/s Compute (per Chip) AMD:~40 PFLOPS FP4 / ~20 PFLOPS FP8 NVDA:~50 PFLOPS FP4 / ~25 PFLOPS FP8 Google: 4.615 PFLOPS FP8 Power (per Chip) AMD: 1000-1400W NVDA: 2300W-3600W(Rubin Ultra) Google: 300W Est. Price AMD: $3.5m-$5m(72 GPUs) NVDA: $8.5m-$10m (144 GPUs) Google: $445m/pod ( 9,216 chips) for 3yrs+ contract Ecosystem AMD: Open-source (ROCm 8) NVDA: Close-source(CUDA) Google: Google Cloud locked ~Overall TCO Winner: AMD MI450 (20–40% Edge in Inference) AMD team emphasizes MI450's "Milan moment" for TCO, driven by TSMC 2nm (20–30% perf/W gain) and chiplets (15–25% power isolation). For memory-bound inference (70% of hyperscaler workloads), Helios delivers 30% more tokens/sec at half the power/rack vs. Rubin. Volume pricing ($30K–$40K/GPU, Meta's 42% allocation) and open ROCm slash soft costs. In a 1 GW cluster, ~NVIDIA Rubin's Strengths (Training Moat, But Higher Costs): Rubin's raw compute (50 PFLOPS FP4/chip) and NVLink (20 TB/s) suit bursty training, but 2,500W-3600W TDP inflates energy require extensive/expensive cooling and CapEx (25–35% premium). CUDA licensing adds ~$4K/GPU/year, and proprietary racks demand retrofits ($1M+ for CDU). TCO balloons 20–30% for inference vs. AMD( 1.13x slower Llama training but 2x power). ~Google Ironwood's Niche (Pod-Scale Inference, But Locked to Google Cloud) , Ironwood's 192 GB HBM3e/chip and SparseCore v3 yield 2x perf/W over Trillium for embeddings/LLMs. But GCP's "pay-per-token" model drives 1.6–2.1x higher TCO due to commitments (~$445M/3 years+ for pod). Pod-scale (9,216 chips) with 10–15% energy overhead and inflexibility for training/multi-vendor setups. Not Financial Advice!

@bjmtweets @MacNeillPatrick Then why not buy at these levels?

@MacNeillPatrick You can compare to $DOCU / ZOOM or $META / $NFLX And I think it's more likely to be $META / $NFLX

$DUOL at $150 is the price that interests me

@ashonzay_7 I agree. That’s 308.64. Thank you

@MPrimgaard Naw I don’t see it stopping, I actually have 268 as the next target and from there $300 .. but all in all I see $AMD pushing for a $500B market cap

$227 PT achieved !!! Next up $250 🎯🎯🎯