MadFan (@vslyhcrypto) header
MadFan (@vslyhcrypto) profile picture

AI || DeAI || RWA || dePIN Idea Generator || AI Agent Creator Content creator Ambassador: @horizon_trade_x @quantpilotapp @opinionlabsxyz partner @partykols

LA
Joined February 2022
Photo by MadFan (@vslyhcrypto)Photo by MadFan (@vslyhcrypto)Photo by MadFan (@vslyhcrypto)Photo by MadFan (@vslyhcrypto)Photo by MadFan (@vslyhcrypto)Photo by MadFan (@vslyhcrypto)Photo by MadFan (@vslyhcrypto)Photo by MadFan (@vslyhcrypto)Photo by MadFan (@vslyhcrypto)Photo by MadFan (@vslyhcrypto)
best Chinese ai company end of august? does someone have doubts who this will be, i don't? unfortunately i can judge about @Kimi_Moonshot only before their K3 release, but i really liked their approach and work with their different versions LLM as soon as they released K3, i of course first thing decided try it, but alas, servers of course overloaded, i understand everything, i got in queue and wait purchase.....but i carefully studied everything that other users shared, showing their results, i won't stop detailed on this, my experience enough visual assessment of model results depending on task i won't talk about numbers of test results, i interested in result that i request and always say and judge exclusively by this parameter and in conclusion say that yes, Kimi K3 will become best Chinese ai company end of august, and maybe not only China and yes, i wait my turn to purchase and will use as main tool in my work in pair with claude, they will confidently complement each other. and maybe i will use only one model and yes, forgot, i always test something new from updated LLM by my certain "task" human, of course, this concerns what this model for and what its function believe my predict, your options you can choose @Polymarket
will we see ai model updates every month? do you want this? i don't think this will happen, this huge money - model training!!! and essentially i understand all development moments, but impression like we entering dead end, this like was with new phone models every year, kind of like.... also i consider that for example closed models get of course profit, but not such gigantic. in turn open models in my view get more open information and ways apply their AI, what adds more value for training materials, and in my view this pretty good way like W-W. well "small" example Qwen3.8-Max and Kimi K3 i long didn't see serious achievements, news and updates about Gemini, except additional tools and they good, so i think next serious release will be from them, from Google prediction i found on my favorite @Polymarket
13
3
33
2,053
MadFan retweeted
got early access to griffin from @tavus this is not a banal "ai chat via video call". you talk to it face to face like a human, and it listens, looks and answers your questions. it interrupts, and lets you interrupt it, without that banal polite turn-taking i asked: "if i start rambling, stop me". waited for the moment and cut in mid-sentence. friends don't do that, but he handled it in less than a minute. in my view that's a cool result at this stage kinda rough, that's the whole thing. and maybe that's the most useful feature i've seen in such projects this year for now it's research preview, you can't build on it. but the direction is clear: we're moving beyond text chats thanks @hassaanraza and team @tavus tavus.io/griffin
Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).
Readers added context they thought people might want to know
The 48% figure and "video Turing test" claim are from Tavus's own study of 54 one-minute calls, not independently verified or using a standard protocol. Griffin-Lite leads NVIDIA's VideoFDB benchmark on their public leaderboard. cellcog.ai/blog/tavus-gri… research.nvidia.com/labs/amri/proj… tech-ish.com/2026/10/02/tav…
5
2
2
19
672
got early access to griffin from @tavus this is not a banal "ai chat via video call". you talk to it face to face like a human, and it listens, looks and answers your questions. it interrupts, and lets you interrupt it, without that banal polite turn-taking i asked: "if i start rambling, stop me". waited for the moment and cut in mid-sentence. friends don't do that, but he handled it in less than a minute. in my view that's a cool result at this stage kinda rough, that's the whole thing. and maybe that's the most useful feature i've seen in such projects this year for now it's research preview, you can't build on it. but the direction is clear: we're moving beyond text chats thanks @hassaanraza and team @tavus tavus.io/griffin
Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).
Readers added context they thought people might want to know
The 48% figure and "video Turing test" claim are from Tavus's own study of 54 one-minute calls, not independently verified or using a standard protocol. Griffin-Lite leads NVIDIA's VideoFDB benchmark on their public leaderboard. cellcog.ai/blog/tavus-gri… research.nvidia.com/labs/amri/proj… tech-ish.com/2026/10/02/tav…
5
2
2
19
672
according to the rumors, the $LAPTOP launch is coming very soon. be extremely careful we’ll probably see a ton of scam contracts and fake tokens appearing in the coming days. I’m fully loaded and ready. I’ll be buying on Fomo (link in comments) planning to buy?
22
3
1
33
4,270
GROK BOT MADE A TRADING DREAM REAL WITH FRIENDS In the summer we went camping with friends and everyone had their own thing, now that’s the strategy I turned my friends into trading agents that make profit. ISLAND - the trading island grill = FLOW, places the order volleyball = FEED, reads the tape kitchen = QTY, runs the models campfire = MACRO, watches the world hookah = lounge where the agents rest on their own boat = Vanya, spins in a circle When the agents are busy, I make money. After a while they go to sleep and in that moment I don’t trade. Every agent is on the job and also has rest time so they don’t overheat. All this became possible thanks to @shubgaur , a guy from the Cursor/SpaceXAI team Also my next goal is for @joshkim to notice me
You can now share templates of your Bots with others.
21
6
1
49
3,025
MadFan retweeted
who will be added to chinese military companies list by june 30, 2027? and this is a very interesting event on @polymarket well, if of course this happens, then it's most likely it will be deepseek although i think anyway that chinese military uses chinese ai, possibly already existing models, possibly their own (which is logical), but i can say that this model should be closed for use in such a sphere i'm just interested where bytedance sneaked in there, with which model? doubao-1.5-pro, seedance😁, trae ai? these are closed models
so, i got it, i'm standing at 40$😄😄😄 while i'll be thinking about it all day, i might as well go to the farmers market in this period, want to buy some fresh vegetables also in my thoughts still get to testing glm, namely their new model 5.3, by tests it beat opus 4.8, kimi k3 and deepseek v4 pro, which is very interesting well and i can't pass by deepseek v4 pro, without noise, just updated the model that caught up with claude at 95%, and costs 45 times cheaper: deepseek v4 pro - $0.44 per million input tokens and $0.87 for output. claude fable 5 - $10 and $50. for the same work the western model comes out ~45 times more expensive i'm looking at how Chinese engineers will soon overtake American models at this pace as i said, we are far from the middle of the path of ai development in general, fasten your seatbelts also i want to look at events on @polymarket and make a review, but that's not for sure, i'll see depending on mood, because i need to finish a few website versions
20
1
33
1,069
MadFan retweeted
for the first time in my life i got lucky, the price of the question was 200$ and i want to give away 10$ right away about how to win 10$, and then i'll tell you about the win what's required: register duelbet.com/c/1azya1 write your uid in the comments so what happened? i sometimes test my luck in various casinos and stuff, not often but it happens the other day i saw a new project for me @duelbetcasino, decided to try, made a deposit of 200$ and was surprised that's not the word, i was blown away, the winnings were 15 000 dollars i didn't think for long what to play and chose the game «zeus» also what's important i got my money without problems + they give bonuses, usually in other casinos there are problems with this maybe luck will be on your side too...
🤝 Paid partnership
12
1
2
38
541
Best Chinese AI Company end of August? Seriously? I don't believe it. Okay, the crowd on @polymarket chose Alibaba as the best one, but I disagree First, as I always say, I don't need weights, I look at the result. It is always subjective and will remain that way. Of course, this applies to people who understand how this whole AI thing works Basically, I think the leaders are Moonshot, DeepSeek, or Zai The point is the result you want to get, not the number of clients who mindlessly use your service And here, of course, the question of development comes up: how do you make people understand that this is actually a tool and not just something to play around with? 100% of the people I know have heard about AI, but they don't understand what it is useful for or how to use it productively. That's the problem All these weights and benchmarks are almost useless to most people
for the first time in my life i got lucky, the price of the question was 200$ and i want to give away 10$ right away about how to win 10$, and then i'll tell you about the win what's required: register duelbet.com/c/1azya1 write your uid in the comments so what happened? i sometimes test my luck in various casinos and stuff, not often but it happens the other day i saw a new project for me @duelbetcasino, decided to try, made a deposit of 200$ and was surprised that's not the word, i was blown away, the winnings were 15 000 dollars i didn't think for long what to play and chose the game «zeus» also what's important i got my money without problems + they give bonuses, usually in other casinos there are problems with this maybe luck will be on your side too...
7
1
16
329
for the first time in my life i got lucky, the price of the question was 200$ and i want to give away 10$ right away about how to win 10$, and then i'll tell you about the win what's required: register duelbet.com/c/1azya1 write your uid in the comments so what happened? i sometimes test my luck in various casinos and stuff, not often but it happens the other day i saw a new project for me @duelbetcasino, decided to try, made a deposit of 200$ and was surprised that's not the word, i was blown away, the winnings were 15 000 dollars i didn't think for long what to play and chose the game «zeus» also what's important i got my money without problems + they give bonuses, usually in other casinos there are problems with this maybe luck will be on your side too...
🤝 Paid partnership
12
1
2
38
541
who will be added to chinese military companies list by june 30, 2027? and this is a very interesting event on @polymarket well, if of course this happens, then it's most likely it will be deepseek although i think anyway that chinese military uses chinese ai, possibly already existing models, possibly their own (which is logical), but i can say that this model should be closed for use in such a sphere i'm just interested where bytedance sneaked in there, with which model? doubao-1.5-pro, seedance😁, trae ai? these are closed models
so, i got it, i'm standing at 40$😄😄😄 while i'll be thinking about it all day, i might as well go to the farmers market in this period, want to buy some fresh vegetables also in my thoughts still get to testing glm, namely their new model 5.3, by tests it beat opus 4.8, kimi k3 and deepseek v4 pro, which is very interesting well and i can't pass by deepseek v4 pro, without noise, just updated the model that caught up with claude at 95%, and costs 45 times cheaper: deepseek v4 pro - $0.44 per million input tokens and $0.87 for output. claude fable 5 - $10 and $50. for the same work the western model comes out ~45 times more expensive i'm looking at how Chinese engineers will soon overtake American models at this pace as i said, we are far from the middle of the path of ai development in general, fasten your seatbelts also i want to look at events on @polymarket and make a review, but that's not for sure, i'll see depending on mood, because i need to finish a few website versions
20
1
33
1,069
so, i got it, i'm standing at 40$😄😄😄 while i'll be thinking about it all day, i might as well go to the farmers market in this period, want to buy some fresh vegetables also in my thoughts still get to testing glm, namely their new model 5.3, by tests it beat opus 4.8, kimi k3 and deepseek v4 pro, which is very interesting well and i can't pass by deepseek v4 pro, without noise, just updated the model that caught up with claude at 95%, and costs 45 times cheaper: deepseek v4 pro - $0.44 per million input tokens and $0.87 for output. claude fable 5 - $10 and $50. for the same work the western model comes out ~45 times more expensive i'm looking at how Chinese engineers will soon overtake American models at this pace as i said, we are far from the middle of the path of ai development in general, fasten your seatbelts also i want to look at events on @polymarket and make a review, but that's not for sure, i'll see depending on mood, because i need to finish a few website versions
I’m raising the bar to $40 – what do you reckon? Come on, pay me for this post🤣🤣🤣
20
3
26
1,346
I’m raising the bar to $40 – what do you reckon? Come on, pay me for this post🤣🤣🤣
My X account is worth $5.02k Is $35 a good rate for a post, or is it low? Show me yours 🫡
11
1
18
599
I WAS RIGHT!GOOGLE HAS RELEASED GEMINI 3.7 FLASH-CHEAPER AND MORE POWERFUL FOR CODING made a quote of my prediction but today i'll try to guess next google gemini pro model released by...? such event is on @polymarket, and i'll give my forecast on this i have just a few options, or rather one or two ~first option is that they won't release pro model at all this month ~second option, august 31, this is basically natural and understandable, either you release a junior model and look at your usage metrics, and then you fix everything that needs to be fixed already in the pro model, and yes, this requires time, which by my logic corresponds to my forecast, launch on august 31. well or you should have released both versions right away, which as we can see didn't happen...
will we see ai model updates every month? do you want this? i don't think this will happen, this huge money - model training!!! and essentially i understand all development moments, but impression like we entering dead end, this like was with new phone models every year, kind of like.... also i consider that for example closed models get of course profit, but not such gigantic. in turn open models in my view get more open information and ways apply their AI, what adds more value for training materials, and in my view this pretty good way like W-W. well "small" example Qwen3.8-Max and Kimi K3 i long didn't see serious achievements, news and updates about Gemini, except additional tools and they good, so i think next serious release will be from them, from Google prediction i found on my favorite @Polymarket
12
2
23
845
IPOs BEFORE 2027? i see an interesting event that is placed on the @opinionlabsxyz platform. and it's interesting because in my view the specified candidates for initial public offering are not quite ready for it, who they are, i'll list some of them: anthropic bytedance canva openai discord i still understand why anthropic, bytedance, openai are on this list, because these are really strong companies in the tech sector, someone of them will have to start trading on exchanges anyway, especially some seem to be ready for this already (openai). real technologies stand behind them, plus, almost everyone knows about their products, the main thing is that the product exists and it works but what are canva and discord doing there, i can't understand, well let's say discord, yes they have their own niche, their own audience and they are really known in different areas. and what canva? what is there in their product that they rushed to ipo in general, i think it will be either anthropic or openai well because honestly i don't see any real competitors and ability to show results in digital products
Team falcons – K27 esports world cup team falcons obvious leader, there's no point arguing here, but they don't always start confidently in big tournaments. plus they had a break. let's see what kind of playing form finn «karrigan» andersen has k27 Russia young collective in this lineup, really want to see what they're capable of and how they'll show themselves against teams of such a level as team falcons. i don't have big hopes on them, but still it's big experience for them more than sure that team falcons will win i find this event and other esports world cup advanced cs2 matches at @polymarket
16
2
23
1,121
Team falcons – K27 esports world cup team falcons obvious leader, there's no point arguing here, but they don't always start confidently in big tournaments. plus they had a break. let's see what kind of playing form finn «karrigan» andersen has k27 Russia young collective in this lineup, really want to see what they're capable of and how they'll show themselves against teams of such a level as team falcons. i don't have big hopes on them, but still it's big experience for them more than sure that team falcons will win i find this event and other esports world cup advanced cs2 matches at @polymarket
OLDBOYS – NEW VISION ESEA ADVANCED CS2 who's gonna win? NEW VISION is going through another difficult period on its way. their recent results show that the team is struggling with adapting to the opponents' playing style, which could become a decisive factor in this match, but they have jame, who sometimes carries the team out of the most difficult situations, but you can't constantly hope for this sadly... OLDBOYS demonstrate strong teamplay, which is confirmed by their high coordination and ability to control the map. over the last five matches they won three victories, which speaks about stability and developed team strategy. also their captain is also a serious leader, which also cannot be discounted so what i want to say, as much as i would like new vision to win, but oldboys have big chances to win due to their stability i find this event and other esea advanced cs2 matches at @polymarket
6
1
11
443
i did something, don't know how useful and in demand it is... true cost check-calculator of real cost of trading on perps @HyperliquidX what was the idea backtest in @quantpilotapp(and almost in any other engine) shows profit by prices. but it doesn't show two things that eat this profit on the real market: !!! exchange commissions - funding rate- the payment that drops every hour between longs and shorts on perpetual futures how to use this you take from your backtest a few numbers: coin, turnover, number of trades, profit, direction of positions. you put it in the calculator. on the output: real profit after commissions and funding breakdown-how much exactly commissions ate, how much funding, with formulas and substituted numbers main number-break-even threshold. how many percent per trade the strategy must earn, just to go to zero. and immediately comparison: fact against threshold why this is not a trifle i ended up like this: backtest +$262 at turnover of $1.4m . real costs on hyperliquid-around $273 (commissions + funding). profit turns into loss-and the backtest simply didn't see this i used quantpilot their quantscript language gives access only to prices and candles- funding rate is physically unavailable there. this is language architecture, not a bug. the platform can't calculate this itself, even if it wants to very much. you have to count outside-this tool does this. it doesn't argue with the backtest, it shows what the backtest by definition doesn't see. and it ends not with a sentence, but with a hint: what threshold the next version of the strategy should pass why i think this is cool? i haven't really seen such calculators for crypto (i might be wrong)-either general estimates without real exchange data, or closed paid tools. here it's the opposite: open source of the calculation core, rates with verification date, funding -from actual hyperliquid history, hour by hour, not averaged. every number in the report is clickable and explains its own formula- not just a number, but transparent logic separate detail: the tool has its own verification page- tests on real numbers, calculated by hand, lie open. a tool that checks other people's calculations, must be checked itself-otherwise it's worth nothing technically all calculations directly in the browser, nothing goes to the server, no wallets and keys. you can copy the result as a picture and share on x, or copy text-task and paste it straight to the quantpilot agent, so he can rebuild the strategy taking into account the threshold who will find it useful anyone who trades perps and wants to know the real price of an idea before launching it on real money-not only to quantpilot users, but in general to everyone who trades perpetual futures on hyperliquid just to give you an idea, I ve included a link in the comments
13
21
1,271
OLDBOYS – NEW VISION ESEA ADVANCED CS2 who's gonna win? NEW VISION is going through another difficult period on its way. their recent results show that the team is struggling with adapting to the opponents' playing style, which could become a decisive factor in this match, but they have jame, who sometimes carries the team out of the most difficult situations, but you can't constantly hope for this sadly... OLDBOYS demonstrate strong teamplay, which is confirmed by their high coordination and ability to control the map. over the last five matches they won three victories, which speaks about stability and developed team strategy. also their captain is also a serious leader, which also cannot be discounted so what i want to say, as much as i would like new vision to win, but oldboys have big chances to win due to their stability i find this event and other esea advanced cs2 matches at @polymarket
So, what's new today? First of all, thanks to @horizon_trade_x for the trust and for choosing me as an ambassador. I'll start working on content today Yesterday I mentioned that I couldn't buy a subscription to @Kimi_Moonshot and get access to K3 because I'm still on the waiting list. That's fine. I took a different route. I sometimes forget about OpenRouter, but I remembered it and topped up my balance for some small tests, which I've already done. Still, I'm waiting to buy the actual subscription because it works differently from API aggregators. The subscription gives you memory and settings, and of course the interface matters too So, what about the tests? Yesterday I wrote about the market on @Polymarket and said that, in my opinion, Kimi K3 should become the best LLM in China. And yes, after a few tests, I still think so. Here's what I did, just some simple tests: ~ I took some old, simple Claude prompts and used them to build several websites in Kimi K3. I have to say I'm more than satisfied with the results. These were prompts originally written for Claude Design. In some cases the results were better, in others they were no worse, even though I didn't refine the prompts or ask for any changes. For now, my conclusion is that once I really get familiar with Kimi K3, the results will be even better ~ Right now I'm building a public service with Claude Code for a project where I'm also an ambassador. At one stage, I decided to double-check Claude Code's prompts with Kimi K3, and the results were excellent. Claude Code accepted 4 changes suggested by Kimi K3, while only 1 remained questionable For me, these two tests say a lot and open up new possibilities that I can already see My conclusion: I believe the combination of Claude Code and Kimi K3 will make me 1000% stronger
13
3
18
1,154
yo guys, i'm involved in this, though i don't talk about it much, i mean about prop firm, but onchain the thing is, i don't trade on my own money, they give it to me when i buy a challenge, and i'm more than happy with it, because i'm not breaking the risks, and that's the main thing in my view i'm not a risky guy, but for such challenges i'm ready, your money stays with you, and mine with me who's the same as me, ready to try without risk, welcome, link in the comments
0/Season 0 is over. Season 1 starts NOW.
9
2
16
1,170
Nikita, thanks for your work. despite the fact that many people reacted negatively to your decisions, you warned about it in advance and sometimes explained what exactly was wrong and unacceptable. the main thing is that you paid attention to this. i found it interesting to analyze your decisions, and i understood your logic p.s. i thought you would also sort things out with articles, posts and then quotes about ai, with people who decided to get monetization that way, who decided to earn that way, especially i’m talking about farmers, but i think you already know that anyway. and i don’t care what haters say thanks
Ladies and gentlemen, it's time to pass the torch and demote myself to my natural state: a poster. I'll be stepping back from leading product for 𝕏 and will continue on as an advisor. Serving the X community has been the privilege of a lifetime. X is, and will remain, the most important communication technology in history. But running this app is a 24/7 job and it's now time for me to take a breather. The app is seeing unprecedented growth in new users & engagement. We continue to break records every month. We've climbed 70 spots in the App Store since this time last year. And in the last 400 days, we rebuilt almost every aspect of X: the Timeline, the Android app, onboarding, notifications, chat and more. We also launched nearly 30 new products while protecting the integrity of the town square: becoming the first app to show Country-of-Origin on profiles and mounting defenses against AI bots. There's certainly much more work to be done, but our foundation is stronger than ever. None of this would have been possible without the incredible team here. The next leaders will take X to even greater heights with @benjitaylor on design, @singhai on core product engineering and @dinkin_flickaa on mobile engineering -- among many other great people. Thank you to Elon and the X team for welcoming me into the company. See you on the Timeline.
3
2
14
1,188