与「2019_New_Years_Day_ISAC🏆」相关的搜索结果

[NOTICE] The moments of athletic gugudan♥ New Year’s Day Idol Star Athletics Championship behind the scenes( ´∀｀) #gugudan# ’s new title as ‘athletic idol’ #2019_New_Years_Day_ISAC🏆# behind the scenes📷 You can check it out from Jellyfish Post now😄 ▶

Our office Wi-Fi password has been the same for six years: "Jessica2019." Every new hire asks who Jessica is. Nobody knows. Except Tom, the IT guy. Tom set it up. Tom refuses to change it. We found out from the receptionist. Jessica is Tom's ex-girlfriend. They broke up in 2019. He set the password the same week. For six years, 80 of us have typed his ex's name about 20 times a day. Last month, Tom handed out wedding invitations to the whole office. We opened them in the kitchen. Then everyone went very quiet. The bride's name was...

Our office Wi-Fi password has been the same for six years: "Jessica2019." Every new hire asks who Jessica is. Nobody knows. Except Tom, the IT guy. Tom set it up. Tom refuses to change it. We found out from the receptionist. Jessica is Tom's ex-girlfriend. They broke up in 2019. He set the password the same week. For six years, 80 of us have typed his ex's name about 20 times a day. Last month, Tom handed out wedding invitations to the whole office. We opened them in the kitchen. Then everyone went very quiet. The bride's name was...

Grok Bot summary: IAC 2026 presentation by SpaceXAI President Michael Nicolls. 🎯 The main message - SpaceX was founded in 2002 to make life multi-planetary and "extend the light of consciousness to the stars." - He said the biggest risk as constellations grow is not debris. It's satellite operators not sharing data with each other. 🌍 How Starlink grew - Starlink began in 2019 with 60 V0.9 satellites. User service started in late 2020 as the "Better Than Nothing Beta." - V1.5 added space lasers, which let the satellites link into a mesh. That's when Starlink really took off, reaching oceans, Antarctica and airplanes. - V2 (2023) has 4x the capacity of V1. Falcon launches V2 satellites 2–3 times a week, and each launch adds about 3 terabits of capacity. - V3 launched on Starship about a week before the talk. Each V3 satellite has another 10x capacity, and each Starship launch carries 20–30x what a Falcon launch does. - Starlink now has over 11,000 satellites, about 60% of all active satellites in low Earth orbit. - It serves more than 13M customers in about 170 markets, plus ships and planes. - The laser mesh means Starlink doesn't depend on local ground infrastructure. During last year's fires and power outages in Spain, it kept tens of thousands of customers and emergency responders online. 🚀 The life of a Starlink satellite - **Before launch:** SpaceX checks that deploying the satellites won't bring them close to anything else, and shares where they'll be. - **Deploy:** satellites are released below 300 km, where almost no other operators fly. - **Orbit raising:** each climb path is checked against other satellites and crewed space stations, and SpaceX publishes the satellites' positions every hour on the way up. - **Lower orbits for safety:** the original plan was about 1,000 km. Now every satellite flies below 500 km, and V3 will fly at about 350–360 km. At these heights a dead satellite falls back to Earth within months instead of staying up for centuries. 🛡️ How they avoid collisions - Each satellite makes small station-keeping moves about once a day and a collision-avoidance move about every 10 days. - Across the fleet, that's about 1,000 avoidance moves a day. - Starlink's threshold for moving is nearly 100x more cautious than the industry standard, so the whole fleet's collision risk is about the same as a much smaller constellation's. - "Ducking" moves turn the satellite so it presents a smaller target during a close pass, cutting the collision risk about 10x more. - Automated software makes these decisions, which is why Starlink tries to stay in contact with every satellite at all times. 📡 Always connected - Starship's live launch video was streamed over Starlink. - The new V3 satellites connected to the network within 2 minutes of release, because Starlink terminals are built into the satellites. Their lasers link into the mesh within tens of minutes. - The goal is to make contact right away and never lose it. 🌬️ Space weather - During a 2024 solar storm, air density in orbit jumped 5x in a few hours, which throws off predictions of where satellites will be. - Starlink is "probably one of the best sensors" for this. It measures density in real time and automatically corrects its position forecasts. 🔭 Working with astronomers - **Optical:** SpaceX tried Vantablack paint and sun visors. What finally worked was a mirror-like film that reflects sunlight away from the ground, combined with careful steering. A V2 satellite now looks about as bright as a sheet of paper. This was done with groups like the Vera Rubin Observatory. - **Radio:** working with observatories like the US National Radio Astronomy Observatory (NRAO), Starlink steers its beams and frequencies around where the telescopes are pointing. As a result, the telescopes can now observe more often. ♻️ End of life - SpaceX uses statistics to predict when parts will fail and deorbits satellites early. Many of those satellites would have kept working. - Retired satellites are steered down within weeks, not left to drift for the 5 or 25 years regulations allow. - They're built to burn up completely, and reentry is aimed over the ocean. - Of about 10,000 satellites launched by early 2026, only a few dead ones remain in orbit. The goal is zero. - "Reliability and rate go together." ⚠️ The biggest danger: other operators - Tracked debris is like obstacles on a road: if you can measure it, you can avoid it. - What you can't avoid is a satellite that maneuvers without telling anyone. Starlink has had passes as close as tens of meters this way. - Starlink has maneuvered around satellites from about 650 operators in recent months, but shares position data with only about half of them. "That number is too low." - SpaceX shares data with China's space station and some Chinese operators. Many operators don't share at all, including some in the US, Europe and Japan. ⭐ Stargaze: Starlink's own space tracking - Over 30,000 star trackers on Starlink satellites spot passing objects as streaks against the stars. - Stargaze now tracks over 17,000 objects, including about 90% of the maneuvering satellites that cross Starlink's orbits. - In one example, an untracked satellite's predicted 9 km miss shrank to about 60 m after it maneuvered toward Starlink. Stargaze caught the move and the Starlink satellite dodged in time. - He said it's still no substitute for operators sharing their plans in advance. 🤝 The ask: share your data - SpaceX runs its own space safety data platform, and over 100 operators have signed up. - Operators can upload their satellite positions and check them against Starlink and against Stargaze tracking. - His message to every operator was to join the platform and share their positions. 🧠 StarMind: AI compute in space - StarMind is SpaceX's planned AI satellite constellation, starting launches and scaling next year. - The goal is to move AI compute off Earth. He expects StarMind to drive demand for Starship launches the way Starlink did for Falcon. - On the ground, SpaceX is building over 2 GW of compute in Memphis. He said AI demand will be nearly limitless, and Earth's resources make that hard to keep scaling there. - In space, the satellites run on solar power and shed heat through large radiators. - Each satellite is simpler than a Starlink: basically a computer with big solar panels and radiators. - Each one has about 250 kW, roughly one NVL72 rack of compute. - They'd fly in tight clusters of about 10 satellites linked by about 10 terabits of bandwidth, then connect to Starlink. - SpaceX has filed for up to **1 million satellites**, in low-inclination and sun-synchronous orbits. Building them could mean thousands of satellites a week. - New challenges include much higher reliability and redundancy, flying in close formation, and micrometeoroids. Dead satellites might be sent away from Earth instead of burning up in the atmosphere. 🌙 Beyond Earth - Starlink is developing very high-bandwidth lasers to connect the Moon, then Mars, building an "interplanetary mesh." - All of that depends on getting safety right in low Earth orbit first. 💡 Key takeaways - Starlink is the world's biggest constellation, and it's moving lower to be safer. - It makes about 1,000 collision-avoidance moves a day, with nearly 100x more caution than the industry standard. - The biggest risk is operators who don't share where their satellites are going. - StarMind could scale to 1 million AI satellites, with launches starting next year. - "Coordination is a key element for realizing all of our ambitions."

NEW: WHISTLEBLOWER 2019 TRUMP IMPEACHMENT - Registered Democrat - "had a prior professional relationship with one of the Democratic Presidential candidates," according to newly released transcripts from the Intelligence Community watchdog Michael Atkinson. In a 2019 briefing to the house intelligence committee, Atkinson added, "I did not find the complainant (whistleblower) was biased." After seven years, the Atkinson records have been finally released @DNIGabbard @RepRickCrawford

SpaceX Starbase: Helping Lift Brownsville Out of Poverty and Bring Wealth to Local Families For many years, Brownsville and the Rio Grande Valley carried the difficult reputation of being among the poorest regions in the United States. High poverty rates shaped daily life for families across Cameron County for decades. In the 1990s, the area was even designated a federal Rural Empowerment Zone in recognition of these long-standing challenges. Today, the trajectory is changing. Poverty rates in Cameron County have been declining steadily. They fell from the mid-30s percent range around 2010 to 28.9% in the 2015–2019 period, and now stand at 24.8% according to the latest U.S. Census Bureau data (2020–2024 American Community Survey 5-year estimates). While the rate is still higher than state and national averages, the consistent downward trend reflects real, measurable progress. A New Wave of Investment and Job Creation Recent data from the Greater Brownsville Economic Development Corporation’s FY 2025 Annual Impact Report shows strong momentum. Between October 2024 and September 2025, the organization helped attract $183.7 million in new investment and supported the creation of 3,288 jobs. The report also shows 10,604 jobs retained during that period and 7,116 jobs already committed for 2026. *Greater Brownsville EDC FY 2025 Key Economic Highlights (see infographic below) These figures reflect broad economic activity across the region, with SpaceX’s Starbase playing a significant role as a major anchor project. Starbase has brought thousands of direct jobs to the area and has helped attract suppliers and related investment. This type of large-scale development is helping address long-standing needs for stable employment and skills development in Brownsville and surrounding communities. Local voices are also noticing the change. Former Brownsville City Councilwoman Jessica Tetreau, @JessicaTetreau at Starbase beach at sunset described the shift she has witnessed in her own community: “Before, in the past in Brownsville, people would talk about the brain drain, how all of our youth would have to leave to San Antonio, to Austin to find jobs… And now these young people that are from the community are finding these amazing jobs.” She shared that in her own neighborhood, parents are now working at SpaceX, and children are growing up excited about launches and rocket engineering, “just like their fathers.” “The kids that are graduating from UTRGV and local universities, they’re coming to work here. It’s really exciting. Engineering is now one of the hottest and exciting things to have in this area.” Direct Support for Education and Downtown Renewal In addition to job creation, there has been targeted investment in the community’s future. In 2021, Elon Musk and the Musk Foundation committed $30 million to the area. $20 million to schools across Cameron County and $10 million for the revitalization of downtown Brownsville. Brownsville Independent School District received more than $2.4 million of the school funding. The money has supported the expansion of Career and Technical Education programs, helping prepare young people for the skilled jobs now available locally. The downtown portion included a $1 million grant to the Brownsville Community Improvement Corporation to strengthen the historic core of the city. Below are a few photos I took while touring downtown Brownsville with Jessica Tetreau right after Starship Launch 12, showing the renewed energy through spots like Main Street Deli, a local bookstore, and Dodici Pizza & Wine. A Community Moving Forward Brownsville has always been a resilient place. In recent years, it has gained access to meaningful new employment opportunities, investment in education, and visible improvements in its downtown. These developments are helping lift families, create local wealth, and support the renewal of the community. Challenges remain, as they do in any place working to overcome long-term economic hardship. But the direction is positive. New jobs are being created, young people are gaining access to better training, and the heart of the city is showing signs of renewal. This is the quieter but very real story of progress happening in Brownsville today that legacy media will keep quiet about.

To everyone in the HTX community concerned about the recent news, I’d like to clarify the situation and reduce unnecessary panic and misinformation. First, this is a relatively common compliance review process. We are actively communicating with all relevant parties to resolve misunderstandings as quickly as possible. HTX also sees this as an opportunity to further strengthen our compliance and risk-control systems, helping the platform become even more resilient. At present: Platform operations are normal User assets remain secure Deposits, withdrawals, and trading are functioning normally There is no need for excessive concern. For users asking what they should do right now, there are two options: 1️⃣ Do nothing You may simply wait while we complete communications with relevant parties and resolve the situation. HTX has operated for 13 years and has weathered multiple market cycles and industry challenges. We remain fully committed to protecting user assets and platform security. Our support team and I will continue to be available 24/7 to assist users. 2️⃣ If you still feel uncomfortable, you may temporarily withdraw assets on-chain Deposits and withdrawals are currently operating normally, and users are free to make their own decisions. ⸻ What happened? On May 26, the UK Foreign Office announced a new round of Russia-related sanctions under The Russia (Sanctions) (EU Exit) Regulations 2019. The list included 18 crypto-related entities and individuals, including a company named “Huobi Global S.A.” According to the UK statement, this entity allegedly provided financial services and technical support to the Russian exchange Garantex and the A7 crypto payment network. However, there is one very important point many people misunderstand: “Huobi Global S.A.” is not the same thing as the HTX exchange platform used by global users today. Many people equate a brand name with a legal operating entity, but global businesses often operate through multiple legal entities across different jurisdictions for compliance purposes. So: Sharing a similar brand name does NOT mean sharing the same legal entity, operational structure, or asset system. The HTX platform used by users today operates independently under its own structure. ⸻ Why users should not panic 1️⃣ The sanctions target a specific legal entity This is not a “brand-wide sanction.” The UK sanctions apply to a specific listed entity and do not automatically extend to all businesses using similar branding. The practical impact is also mainly limited to the UK financial and regulatory system. ⸻ 2️⃣ The sanctions mainly affect relationships inside the UK financial system This may include: Restrictions involving UK financial institutions Suspension of payment or intermediary relationships Asset-related measures within UK jurisdiction But this does NOT mean: Global user assets are frozen HTX has stopped operating Users cannot trade or withdraw funds At this time, the platform continues to operate normally. ⸻ Why did the situation escalate so quickly? Some third-party blockchain security providers applied broad risk labels to related wallet addresses in a “one-size-fits-all” manner. This affected certain normal user transactions and created unnecessary panic and speculation. Our compliance, security, and legal teams are already communicating with the relevant parties, and we expect the issue to be resolved soon. We understand the community’s concerns and will continue to communicate transparently. If you have any questions, please feel free to reach out to us or our support team anytime.

Here's the #1# thing most people don't know about Warren Buffett: There is nothing special about Buffett’s stock picking. That doesn’t mean that Buffett wasn’t a great investor. He was! Buffett was, by far, the greatest investor in history, by a huge margin. Over 486 months between October 1976 and March 2017 –— 41 years –— Berkshire Hathaway’s Class A stock earned an average excess return of 18.6% per year above U.S. Tbills. Annualized volatility was 23.5%. Sharpe ratio: 0.79. Berkshire’s Sharpe ratio of (0.79) is roughly 1.6x times the broad U.S. stock market’s Sharpe ratio of 0.49 over the same period. Among all large-cap U.S. stocks and mutual funds with 30-plus-year continuous track records, those are unmatched numbers. A dollar invested in Berkshire on October 31, 1976, was worth more than $3,685 by March 31, 2017. A dollar invested in the S&P 500 with dividends reinvested over the same period was worth approximately $76. Buffett beat a passive index by a multiple of 48. But he didn’t do it with stock picking! Three researchers at AQR Capital Management –— Andrea Frazzini, David Kabiller, and Lasse Heje Pedersen –— dissected Berkshire’s 50 years of investments through 2013. They expanded and republished their findings in 2018 in the Financial Analysts Journal, which is the most highly respected industry financial journal. Their work won the Graham and Dodd Award for the best published paper of the year. The paper is called Buffett’s Alpha. They found, after accounting for cheap leverage (from the insurance float) and exposure to a handful of publicly documented factor premiums, Buffett’s investment skill –— the portion of his returns that cannot be explained by any mechanical strategy –— is 0.3% per year. That's statistically indistinguishable from zero. In other words, the alpha that Berkshire enjoyed for 50 years (as it compounded capital at 24% a year!) wasn’t due to Buffett’s stock picking. So, how did he do it? He did it by gaining access to a huge amount of investment capital that he did not own, for free. Buffett’s track record was built on leverage. That’s a dirty word for most investors, but it's the secret behind Berkshire. The AQR researchers had access to something most Buffett commentators do not: 40 years of Berkshire’s audited financial statements and the full quarterly history of the public 13F stock portfolio. The researchers asked a specific question: If I take Berkshire’s monthly stock returns from October 1976 through March 2017, and I run a linear regression against a set of well-documented risk factors –— market beta, size, value, momentum, and two newer factors called Betting-Against-Beta and Quality-Minus-Junk (detailed below) –— how much of Buffett’s performance can the factors explain? And after the factors have been stripped out, how much excess return remains? The data show clearly there are a few qualities that drove Berkshire’s results. First, Buffett has always preferred large-cap stocks, contrary to the popular image of him as a small-cap value investor. He buys elephants. Second, no surprise, Buffett buys cheap. Berkshire is almost six standard deviations away from neutral on the value axis. So far the picture is ordinary. Every large- cap value manager in America loads positively on size and on value. Buffett’s genius lies in the last two factors. These last two factors are a little complicated, but please stick with me. There’s a new factor, that, like value and size, characterizes Buffett’s strategy. It’s called Betting-Against-Beta (“BAB”). What it means is intentionally investing in stocks with very low volatility. The BAB factor captures the excess return that accrues to investors who own low-beta stocks. Low-beta stocks have historically earned higher risk-adjusted returns than high-beta stocks. Financial theory teaches that higher beta (higher risk) should mean higher return. But it doesn’t. The opposite occurs, in fact. And Buffett was one of the very first people to figure this out. Why does this factor persist? In an efficient market, once that factor is known to investors, then they should bid the price up on low- beta stocks until it no longer provides an edge. The explanation, per the theory of AQR’s Frazzini and Pedersen’s theory, is that because ordinary investors do not use leverage and seek high returns, they create persistent excess demand for more volatile stocks. (Having worked with retail investors for 30 years, I can assure you that is true.) But, an investor with access to cheap leverage –— Warren Buffett, for instance –— can exploit the mispricing by owning the low-beta names and levering them up to produce market-beating returns. And the last factor that matters to Buffett is quality. Buffett buys companies with high returns on invested capital. Quality-Minus-Junk (“QMJ”) is a factor described by Cliff Asness, also at AQR with Frazzini, and Pedersen, in a 2019 paper in Review of Accounting Studies. The QMJ factor captures the return to owning stocks of high-quality companies –— profitable, growing, safe, with high payout ratios –— against stocks lacking those characteristics. QMJ has been positive and statistically significant in every major developed equity market for which it has been measured. Berkshire’s loading is 0.37, with a t-statistic of 4.6. –– meaning it is highly significant to Berkshire’s results. In plain English: Buffett only buys large, high- quality, low-volatility stocks of the highest quality. But, Berkshire’s results were not, in any way, unusual. Any investor buying these same kinds of stocks would have earned those same returns –– about 16% a year over time. So how did Berkshire compound at 23% a year? To figure that out, AQR’s researchers built a Berkshire replica. They constructed a simple, rules-based, publicly investable portfolio that mechanically tilts toward large-cap, cheap, low-beta, high-quality stocks, and levers it 1.6- to- 1 to match Berkshire’s insurance float leverage. The correlation between their replica’s returns and Berkshire’s were virtually identical. The authors’ conclusion is unambiguous. “In summary, we find that Buffett has developed a unique access to leverage that he has invested in safe, high-quality, cheap stocks and that these key characteristics can largely explain his impressive performance.” Berkshire’s cost of insurance float has averaged almost three percentage points below the Treasury bill rate across 50fifty years of data. In roughly two-thirds of all years, Berkshire has been paid to hold other people’s money. That is not an investment strategy. That is a financing miracle. It is also the living, breathing heart of Berkshire Hathaway. It’s what Buffett built, starting in 1967 when he paid $8.6 million for National Indemnity’s $19.4 million of float. And it is the factor every retail investor admiring Berkshire’s returns has never paid any attention to. The 1.6-to-1 leverage that AQR measured over the full period, financed at this negative cost, explains the dollar magnitude of Berkshire’s returns. How do we know? An unleveraged version of the same stock portfolio –— which you can approximate by looking at the 13F holdings alone –— has earned an average excess return of 12% percent per year. It’s Berkshire’s leverage that magnifies this excess return to 18.6 %percent. How does this square with Berkshire’s reported gains? Berkshire’s 18.6% excess return, plus the T-bill rate that averaged roughly 4.7% over 1976–2017, gives you a total nominal return of roughly 23% per year, which is the figure you usually see quoted for Berkshire’s historical performance. The 23% tells you what Berkshire returned. The 18.6% tells you how much of that return was compensation for taking investment risk, as opposed to the baseline yield every lender to the U.S. government was earning anyway. With both of Berkshire’s “edges” –— systematic factor exposures to cheap, high-quality, low-volatility stocks and roughly 1.6-to-1 leverage delivered with insurance float –— you get Berkshire Hathaway’s 23% annual gains over 60 years. It’s the structure that’s genius, not the stock picking. And that's very important because it means the original Berkshire formula can work for any investor. I show you exactly how, in my new book.

"I’ve weathered the storm, so let me hold the umbrella for you." Those with scalp issues can continue reading; others may skip. Eleven years ago, I joined the crypto world and began a life of chronic sleep deprivation. After trying a new shampoo, I started experiencing recurring large flakes of dandruff that fell off in chunks, as if my scalp were peeling. But I was too busy to care and occasionally used **Ketoconazole shampoo** (Figure 3). Years later, during a nomadic phase, I randomly switched shampoos. By 2018, my scalp was covered in painful bumps, and the itching became so severe it disrupted my sleep. In 2019, I finally saw a doctor, who prescribed **Selenium sulfide shampoo** (Figure 2). My condition fluctuated until I found lasting relief with long-term use of the **yellow-bottle shampoo** and **Vichy Dercos Green Clay Shampoo** (Figure 1). Compared to others, the Vichy shampoo has a milder medicinal scent, so I highly recommend it. Looking back, the real issue was pushing my body to its limits, delaying medical care, and ignoring its distress signals. I hope everyone prioritizes their health and seeks help early when problems arise. Stay well! PS：The picture comes from the Internet 淋过雨，所以给你撑把伞，如果你头皮有问题的继续，没有的可以跳过。 十一年前加入币圈，开启熬夜生涯，在一次用了一个新的洗发水后开始反反复复起大块头屑，一块块的掉像是头被剥皮。 但因为太忙不以为意，买了采乐（图三）偶尔用，几年后因为流浪地球就很随机的用洗发水，18年极端的时候满头包，痒起来睡觉都睡不好，19年才开始看医生，医生开了二氧化硒洗发水（图二）；后来时好时坏，最终在黄瓶和薇姿绿标洗发水（图一）的长期加持下摆脱痛苦，对比起来绿瓶的洗发水药味较浅，亲测有效，给大家种草。 回头看这十年身体透支，不及时看医生，忽略身体求救信号才是问题，祝大家都身体健康，有问题及时看医生。 图片来自网络，非广告。😆

In 2019, OpenAI announced GPT-2 with this post: Today (~5 years later) you can train your own for ~$672, running on one 8XH100 GPU node for 24 hours. Our latest llm.c post gives the walkthrough in some detail: Incredibly, the costs have come down dramatically over the last 5 years due to improvements in compute hardware (H100 GPUs), software (CUDA, cuBLAS, cuDNN, FlashAttention) and data quality (e.g. the FineWeb-Edu dataset). For this exercise, the algorithm was kept fixed and follows the GPT-2/3 papers. Because llm.c is a direct implementation of GPT training in C/CUDA, the requirements are minimal - there is no need for conda environments, Python interpreters, pip installs, etc. You spin up a cloud GPU node (e.g. on Lambda), optionally install NVIDIA cuDNN, NCCL/MPI, download the .bin data shards, compile and run, and you're stepping in minutes. You then wait 24 hours and enjoy samples about English-speaking Unicorns in the Andes. For me, this is a very nice checkpoint to get to because the entire llm.c project started with me thinking about reproducing GPT-2 for an educational video, getting stuck with some PyTorch things, then rage quitting to just write the whole thing from scratch in C/CUDA. That set me on a longer journey than I anticipated, but it was quite fun, I learned more CUDA, I made friends along the way, and llm.c is really nice now. It's ~5,000 lines of code, it compiles and steps very fast so there is very little waiting around, it has constant memory footprint, it trains in mixed precision, distributed across multi-node with NNCL, it is bitwise deterministic, and hovers around ~50% MFU. So it's quite cute. llm.c couldn't have gotten here without a great group of devs who assembled from the internet, and helped get things to this point, especially ademeure, ngc92, @gordic_aleksa, and rosslwheeler. And thank you to @LambdaAPI for the GPU cycles support. There's still a lot of work left to do. I'm still not 100% happy with the current runs - the evals should be better, the training should be more stable especially at larger model sizes for longer runs. There's a lot of interesting new directions too: fp8 (imminent!), inference, finetuning, multimodal (VQVAE etc.), more modern architectures (Llama/Gemma). The goal of llm.c remains to have a simple, minimal, clean training stack for a full-featured LLM agent, in direct C/CUDA, and companion educational materials to bring many people up to speed in this awesome field. Eye candy: my much longer 400B token GPT-2 run (up from 33B tokens), which went great until 330B (reaching 61% HellaSwag, way above GPT-2 and GPT-3 of this size) and then exploded shortly after this plot, which I am looking into now :)