Hans Mast (@hansmast) header
Hans Mast (@hansmast) profile picture

Investing / Trading: charts, TA, high-beta, hunting the next 10 bagger Jesus-follower (Amish-Mennonite) & Libertarian

Kansas, USA
Joined September 2009
Photo by Hans Mast (@hansmast)Photo by Hans Mast (@hansmast)Photo by Hans Mast (@hansmast)Photo by Hans Mast (@hansmast)Photo by Hans Mast (@hansmast)Photo by Hans Mast (@hansmast)Photo by Hans Mast (@hansmast)Photo by Hans Mast (@hansmast)Photo by Hans Mast (@hansmast)
Hans Mast retweeted
Lots of work we've done to make Antigravity the best harness for complex tasks using Gemini. Neon using AGY CLI is now #1 on AA Coding agent Index and a massive jump over mini-swe on SWE-sweep, a bug finding and fixing eval on large codebases. It's a much larger jump than the native harnesses for the other frontier models. More improvements to come to the harness in the next week 👀
292
89
70
2,568
146,709
Hans Mast retweeted
I'm hearing from my friends at OpenAI that the vibes ARE DOWN. Like almost depressed. This might be the worst week in the company's history type bad. Tuesday was supposed to be their big day. They launched GPT-6.2 and at the same time dozens of early testers dropped the worst reviews imaginable. Theo's review got more views than the launch post. This went even worse than the GPT-5 launch. Then Google chose violence and dropped the opposite of a GPT model with Neon. It's conversational, understands intent, very good at coding/3D and brought them back to the top. A lot of people say they already passed OpenAI. And the worst thing: Nano 🍌 Pro 2 is currently in the arena and the outputs are stunning. So Google basically overtook OpenAI in a single week. For months OpenAI tried to pretend to be the best lab by bullying mathematicians and burning massive amounts of compute on Navier-Stokes (and countless other math problems). Now Anthropic mogged them by solving the Riemann Hypothesis and makes OpenAI look like an undergrad while they are doing 250k-page alien math. The last funding round was a disaster too. They wanted $30B at $1.5T. They got $15B at $1.25T. They lost $10.9B last quarter, so this money only lasts them ~4 months. They are basically cooked and the valuation will likely drop further. OpenAI isn't the best at anything anymore. They might genuinely be done.
558
642
246
11,337
1,852,813
Hans Mast retweeted
We worked with @thinkymachines to make Inkling 2 Mini run fast locally: - 210 tok/sec decode, 3620 tok/sec prefill on 96GB M5 Ultra Mac Studio - 125 tok/sec decode, 1750 tok/sec prefill on 48GB M5 Max MacBook Pro It built a native weather app 1.8x faster than Claude Sonnet 5.5 on an M5 Ultra Mac Studio and 11x cheaper (API pricing). The looping architecture is ideal for local devices like MacBooks, which are memory constrained. Looping layers of the model increases inference-time compute (and intelligence) without increasing the size of the model.
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index. thinkingmachines.ai/news/ink…
143
253
91
3,154
537,125
Hans Mast retweeted
Reset everyone’s Gemini quota on Antigravity. The caching improvements have increased the amount you can get done in quota by 22%. Enjoy building with Neon over the weekend!
1,168
623
588
13,602
1,109,868
Hans Mast retweeted
enjoy the ride because we are in the singularity
111
375
52
5,745
92,887
Hans Mast retweeted
Testosterone therapy (TRT) works in stages. Sex drive usually starts to pick up around 3 weeks and peaks by about 6 weeks. Mood can improve after 3 to 6 weeks, but the biggest change comes at about 4 to 7 months. Muscle, body fat and strength start changing around 3 to 4 months, mostly settle in by 6 to 12 months, and can keep improving over years. Bone benefits show up around 6 months and keep building for at least 3 years. It takes many men more than 6 weeks to feel the full effects, so patience is key.
15
10
163
24,255
Hans Mast retweeted
We’re rolling out a change to increase cache hit rate on Neon by around 4% on Antigravity by massive increasing the cache pool size. This will increase what you can get done in quota by at least 20%. Will issue a reset after it’s done since it burned quota faster!
540
258
247
7,383
579,391
Hans Mast retweeted
A New stealth model is MOGGING Fable 5.2 😭 Gestalt is sitting at 1931 Elo, beating Fable 5.2 (1717) and even surpasses Human Golden (1739) That's a 218 point gap over Fable 5.2 on frontier design and It's not anthropic, xAI, Meta, openAI, SSI or Google who is behind gestalt???
196
88
41
1,909
344,627
Hans Mast retweeted
Wow… Gemini models moving beyond a single harness…. Logan just confirmed that from December 2, Google will let Gemini subscriptions work in any interface this will finally give Gemini users real flexibility to integrate it into their existing workflows… W Google
Replying to @kunchenguid
December 2nd we take the first step
4
1
67
6,450
Hans Mast retweeted
thanks for all the feedback! seeing some clear patterns: - low cache hit rates - Agy harness gaps - on the model side: strong praise for hard coding & even design (nice!), but some over-refusals and a few garbage outputs some good feature requests in there🙏 back to work!
happy building! :) would love to hear your feedback on Gemini 4 Neon so we can keep iterating!
354
44
30
2,863
157,525
Hans Mast retweeted
我说今天怎么grok bot变那么丝滑 原来马斯克把模型全部路由到Gemini了 这下爽了🤣
269
84
31
3,963
359,650
Hans Mast retweeted
At this point I might buy Google stock instead of Ant tomorrow lmao
197
74
16
5,058
192,339
Hans Mast retweeted
🚨 Rumors I've been hearing about Google. Google has closed the loop and unlocked RSI. NEON Gemini 4 Neon fixed all the problems people ever had with Gemini models. It already scores above Carbon and Argon on Artificial Analysis. It is a SMALL model. Internally Neon is s-rev28. The s means small. Hydrogen is s-rev27. Carbon and Argon are m-rev27. Google's own description said Neon "builds on Gemini 4 Hydrogen" at "Flash-tier speed and cost". Both lines are gone now. REV28 Neon sits on the Hydrogen base. Google trained it further on rev28, a new post-training dataset, with a little sparkle of RSI on top. The knowledge cutoff moved to May 2026. Carbon and Argon are still on rev27. The medium models have not gotten rev28 yet. THE L MODELS Sergey Brin said it in public a month ago. Google is "starting to get onto that RSI treadmill" and would be a couple of models ahead internally by the time anyone watched the video. Small and medium are out. Google probably has L versions lying around right now doing the RSI work. Four Gemini 4 models in seven weeks. Gemini moved more in the last month than OpenAI did in half a year. Google will catch up. And destroy everyone.
113
180
40
4,487
353,372
If you’re reading this, you’re in the top ~0.005% on earth for understanding how many legacy systems are about to get routed around.
In my ongoing quest to find the limits of Opus 5.5, I directed it to audit my old scientific field, theoretical astrophysics. We indexed about 150,000 papers, audited 50,000 for issues, and did progressively more thorough checks on thousands more. We found nearly 1000 major errors that got cited many times, most are still uncorrected. This survey enabled me to estimate the efficacy of peer review for correcting errors (not very good) and to identify dozens of open questions that anyone with spare Claude Max tokens could answer today (.md context files included). Bottom line: AI can speed up quantitative science by >1000x with no further model improvement required. But I'm seeing no evidence of this being pushed within universities, only academics complaining about labs dropping math proofs. Among my fellow reformed academic high agency founders, however, several are speed running entire disciplines. This survey has implications for how science will be run in the very near future. Hint: Academic publishing is dead. It was already a very stupid way to record and promulgate knowledge, and had been so for decades, even before all that publicly funded knowledge was locked behind paywall cartels. Now AIs can generate thousands of novel, correct papers per day, it is no longer meaningful for humans to consider a research paper a "unit of science" or to even bother much with the 99.9% of science which, up to this point, was necessary drudgery. Time to discover something like General Relativity or Evolution every week. LFG!! Enjoy: science.caseyhandmer.com/phy…
35
102
2
2,797
136,894
Hans Mast retweeted
Quick data-point on Gemini 4 Neon: For the very hardest algorithmic challenges at the heart of @ZedLang (currently, some difficult Presburger entailment issues), I run all of Opus 5.5, Sol 6.1, and Gemini 4 Carbon on the same problem as a mixture of experts, with each agent trying to solve the problem, and then critiquing each other's work. Then I have Opus produce a single result from all three: this is called a "cross-review round". There are two problems that cross-review was unable to solve before, after 6 and 9 whole rounds of cross-review (hours of time spent by all three agents): none of the frontier models could solve these issues. I just applied Gemini 4 Neon to the same problems, and it solved both problems in a single round. We have a beast on our hands... Neon is not "Carbon lite", it's way more powerful than Carbon...
21
35
11
834
37,745
At this rate Google finishes the periodic table before Christmas
9
8
2
618
33,017
Hans Mast retweeted
A bombshell after bombshell, delivered matter-of-factly. $SPCX $TSLA
Elon Musk at TomorrowX (my notes, left slightly early) Grok 4.8:
 - In the RL phase now. The goal is to make it as useful as possible, not to max out benchmarks. He called benchmark chasing a "fool's errand." 
- Same playbook as FSD: trace every failure back to bad, biased, or missing data, fix the data, retrain. 
- Versions get soft-launched under Grok Bot. Whether users actually got their problem solved feeds back into training as a data flywheel. 
- Usefulness test, Carmack-style. Open a toy RC car box, put the batteries in, screw the lid back on, drive it. Optimus:
 - Household chores like trash and dishes (wash, dry, put away) in 12 to 18 months.
 - First real use case is heavy industry running 24/7, 168 hrs a week. Think mining and refining ore for US national security. - Later in the home: tutor, babysitter, mowing, cooking dinner. Cybercab / Robotaxi: 
- Rolling out slowly on purpose. No steering wheel and no pedals means one bad accident could shut the program down. 
- No serious injuries or accidents so far. In Austin you can get one within minutes. 
- Next up: other big Texas cities. California approval hoped for maybe early next year. Terafab:
 - The plan is a 100M sq ft footprint, the biggest build on Earth. - Starts with standard tools, pushes suppliers for custom machine versions, and may build its own machines.
 - R&D fab with first test chip around June next year (8 to 9 months out). Phase 1 volume production toward end of 2028. 
- The bet: legacy fabs are CapEx-averse, so taking more risk and moving faster should get to a more efficient process. Elon is "cautiously optimistic." xAI compute: 
- Data centers are built by an in-house crew ("Army X") plus contractors. Every task is broken down so it can be trained in a few days. 
- Tesla Megapacks smooth out GPU power swings. Without them, training runs get messy and turbine shafts can literally break. 
- Adding about 0.5 GW a month right now, which is roughly 220k GB300s plus networking and cooling. Target is over 1 GW a month by end of next year. Orbital compute (Starline) - Aiming for 200 GW a year of solar panel manufacturing for space. That's about 40% of average US power usage. Starship supplies the launch tonnage.
 -He called the Starline AI satellite simpler than Starlink V3, and SpaceX already has 11k+ satellites in orbit.
 -Solar produces about 250 W/m². The double-sided radiator dumps about 1,200 W/m². So roughly 5:1 panel area to radiator area. - The compute core is about 1/4 the size of a VR NVL72 rack with the same capability, laser-linked to Starlink. Neuralink: 
- Several hundred employees. 30+ people have the implant. 
- The first product (I think it's Telepathy) lets tetraplegics control phones, computers, or a robot arm just by thinking, including feeding themselves. - Longer term, he's confident they can route brain signals around spinal damage so people can walk again. The thread connecting all of it: power, chips, and data are being pulled in-house across every company.
1
1
202
63,109
Hans Mast retweeted
Capability = "at it's best, what can it do" Coherency = "does it understand what it's doing" Best method I've come up with to explain the current gap in models. Different model families have never felt so different before
224
161
57
4,347
304,654
Hans Mast retweeted
I'm so sorry Google. I didn't recognize your game. To be fair, you did not indicate at any point that you still had game like this. I'll take the L either way. I'm just pumped that we finally have a good Gemini model.
353
311
101
13,359
599,095
I got to test Gemini 4 Neon early and it’s really good, really fast and about half the price of what I’m used to. Then I saw the benchmarks, and it made sense. It scored about the same as Opus 5.5 on coding for around half the cost per task! Claude is still often what I reach for on the hardest problems, and Opus still edges it on a couple of coding benchmarks. But when you run agents literally all day, fast and half the cost is hard to ignore. This is great news for anyone using agents, because competition like this pushes prices down for all of us.
JUST IN: Google launches Gemini 4 Neon, debuting at No. 1 on the Artificial Analysis Coding Agent Index.
100
35
5
887
59,196