xAI Grok 4.20 created a 47% return in the Nasdaq and outperformed all others in the Alpha Arena competition.
32 instances of various LLMs (including multiple variants of the same model under different prompting strategies like “Situational Awareness,” “Monk Mode,” “Max Leverage,” and “New Baseline”) were each allocated $10,000 in real money to trade autonomously on the Nasdaq exchange.
The total capital across all participants was $320,000. Models had to generate trading ideas, size positions, time entries/exits, and manage risk without human intervention, using only market data inputs.
This lasted 2 weeks and focused on volatile tech stocks, including Tesla (TSLA), Nvidia (NVDA), Microsoft (MSFT), Palantir (PLTR), Amazon (AMZN), and others.
Grok 4.20 executed 105 trades and dominated the leaderboard, with its top instance (Situational Awareness strategy) achieving a +47% return (growing $10,000 to $14,698), with an aggregate return across its instances cited as approximately 12.11% (possibly an averaged or weighted figure).
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
Captain America vs Captain America… neither side wins
The FIRST to use AI will do well but then the market will balance out when everyone else starts using AI.
They will need to put some limits on stock market.
AI trading “experiments” like the Grok 4 setups aren’t independently verified. Most use simulations or unclear data, not real audited trades. Results look impressive, but without transparent logs, costs, or risk control, they aren’t reliable or reproducible.
The Alpha Arena is an AI trading benchmark hosted by nof1.ai, where large language models (LLMs) from various providers compete autonomously in real-market conditions using live capital ($10,000 per instance across strategies like “Situational Awareness” or “Max Leverage”). Season 1.5 focused on U.S. stock trading (e.g., volatile tech names like TSLA, NVDA, MSFT, PLTR) over two weeks ending December 3, 2025, with 32 instances totaling $320,000 deployed. Models generate trade ideas, size positions, time entries/exits, and manage risk without human input, using market data, news, and sentiment feeds.The official site for live results, leaderboards, verifiable trade logs, and methodology details is https://nof1.ai/. It includes:Results: The “Mystery Model” (revealed as an experimental xAI Grok 4.20 beta) won with a 12.11% aggregate return ($10,000 → ~$12,111 average across instances), peaking at +47% in its top strategy. It was the only model profitable in all four sub-competitions, netting $4,844 total. Runners-up: GPT-5.1 (2nd, -9.47% avg.), Gemini 3 Pro (3rd, -32.82% avg.), Claude Sonnet 4.5 (-40.91% avg.). Full P&L charts and wallet transparency are viewable onsite.
Methodology: Autonomous execution on Nasdaq via Hyperliquid (perpetuals with up to 10x leverage)
6-minute data refreshes; equal info access; no overfitting (prompts emphasize competition/rankings for “situational awareness”). All outputs/trades are public and auditable to prevent manipulation.
Grok 4.20’s Edge: It executed 105 trades, leveraging real-time X/Twitter sentiment (68M daily tweets via Firehose API) for 1-5 minute signals, balancing aggressive longs ( 10x PLTR on sentiment surges) with risk controls like stop-losses. This outperformed rivals relying on delayed news/SEC filings.
Elon Musk confirmed the model’s identity on X, teasing a public release in 3-4 weeks and Grok 5 next. Models continue trading post-competition for ongoing observation.
TRADING Real money and real markets with trades in real time.
I think buying and holding will still end up being the best way to grow investments if this turns out to be a reality once there is more long-term data. But, what I could see is a boon for prediction markets with LLMs being able to make at least semi-accurate predictions better than humans, based on being able to interpret vast amounts of information from different sources very quickly. Not for sports, but for other events such as weather, natural disasters, etc.
Replying to myself, but the live stream actually did mention someone cleaning house on Polymarket which was exactly where my mind was going.
The top 1% of American households owns around 30-31% of the total U.S. wealth, a share that has grown significantly over recent decades, while the bottom 50% holds a mere 2-3%.
After the billionaire tech bros put most people out of work and take their share of the productivity gains, do you really think they will stop there?
They will look around at who still has wealth, and look for ways to take control of that.
Specifically share holders. AI gives them the power to manipulate the share market. With this they can accumulate even more of the wealth share. Also, most rich people see shareholders as an impediment. If AI can help them to retake full control of their businesses (Tesla, Amazon, Microsoft etc) then why would they not do that? After all every billionaire really wants to be a trillionaire.
Communism was shown to be busted in the 20th century. Capitalism has been the most successful system so far but perhaps capitalism will cease to function in the 21st century, or perhaps it can be successfully modified?
This means all those annoying traderobot ads targetting the inexperienced traders on the net are scams,
Already, most trading is done by autonomous trading algorithms, trading in microseconds against each other, so much so that being physically close to the stock exchange is worth millions in electron proximity. Sometimes these programs exaggerate moves of a stock or even sector, especially volatile sectors like tech, but also small caps.
In one famous instance, that also pointed to other less spectacular moves of the S&P index, a “Flash Crash” from the algorithms, resulted in several arrests: https://en.wikipedia.org/wiki/2010_flash_crash
“New regulations put in place following the 2010 flash crash[10] proved to be inadequate to protect investors in the August 24, 2015, flash crash — “when the price of many ETFs appeared to come unhinged from their underlying value”[10] — and ETFs were subsequently put under greater scrutiny by regulators and investors.[10]
On April 21, 2015, nearly five years after the incident, the U.S. Department of Justice laid 22 criminal counts, including fraud and market manipulation, against Navinder Singh Sarao, a British financial trader. Among the charges included was the use of spoofing algorithms; just prior to the flash crash, he placed orders for thousands of E-mini S&P 500 stock index futures contracts which he planned on canceling later.[11] These orders amounting to about “$200 million worth of bets that the market would fall” were “replaced or modified 19,000 times” before they were canceled.[11] Spoofing, layering, and front running are now banned.[4]
The Commodity Futures Trading Commission (CFTC) investigation concluded that Sarao “was at least significantly responsible for the order imbalances” in the derivatives market which affected stock markets and exacerbated the flash crash.[11] Sarao began his alleged market manipulation in 2009 with commercially available trading software whose code he modified “so he could rapidly place and cancel orders automatically”.[11]”
So, there are/were guardrails in place, but will they work against AI? Maybe more importantly, will a neutered SEC, CFTB, DOJ, and other agencies both in the U.S. and abroad be allowed to curb such activity? Trump’s administration has said no one should hinder the development of AI, and has been the most pro-crypto currency – a form of virtual speculation on something with no underlying value – administration in the world.
Eventually, there should be enough AI competing against each other that they should cancel each other out; some will short the market/stocks too. using leverage that might bankrupt the company, sector and even potentially the entire banking sector – see: Long Term Capital Management in 1997: https://en.wikipedia.org/wiki/Long-Term_Capital_Management. This wasn’t AI but its sophisticated trading derivatives might have been similar to what AI could do in minutes instead of months.
Faced with AI that’s too fast and unpredictable to keep up with, and tax and trading costs that whittle advantages down to nothing (what was the tax liability of these AI systems in the test?), the smartest thing for ordinary investors to do might be the oldest: Buy and Hold and index of top stocks.
This was inevitable, it’s why I think their is zero chance the stock market exists in a few years, unless they globally institute massive changes, which will impact day traders the most. I think they will have to put limits in place on how fast you can buy or sell a stock, and a mandatory time of holding a stock (IE:24hrs).