← Tech Brew Ride Home
Tech Brew Ride Home
What If GPT-5.5 Is Actually Way Ahead?
00:21:08
Robinhood launched agentic stock trading, letting users link Claude or Cursor to dedicated accounts. Micron hit $1T market cap in record time on AI memory demand. YouTube now auto-labels AI content, a new coding benchmark crowns GPT-5.5 the clear leader, and Roku overhauls its homescreen.
Loading summary
Transcript8 lines
- [00:00]
A
Study and play come together on a Windows 11 PC and for a limited time, college students get the best of both worlds. Get the unreal college deal Everything you need to study and play with select Windows 11 PCs. Eligible students get a year of Microsoft 365 Premium and a year of Xbox Game. Pass ultimate with a custom color Xbox wireless controller. Learn more@windows.com studentoffer while supplies last ends June 30th terms at aka mscollegepc.
- [00:35]
B
Welcome to the tech we write home for Wednesday, May 27, 2026. I'm Brian McCullough. Today, Robinhood launched Agentic stock trading, letting users link Clauder Cursor to accounts. Micron hit $1 trillion in market cap in record time. YouTube now auto labels AI content, a new coding benchmark crowns GPT 5.5, the clear leader, and Roku overhauls its home screen. Here's what you missed today in the world of tech. From the Agents for Everything file. Robinhood has launched a feature to let users link AI agents such as Claude or Cursor to separate dedicated investment accounts for trading stocks autonomously, Quoting the journal Robinhood is part of a crowded field of financial firms that have introduced AI tools into a range of services they offer individual customers, from stock research to automated investing management. The agentic trading and credit card spending feature marks another step forward, said Abhishek Fettapuria, Robinhood's vice president of product management. These AI agents for consumers have started to trade in the market, fettapuria said. One thing that we've learned from talking to our customers is that they want to give their agents the power of Robinhood, but in a very safe way. Robinhood users can link an AI agent like Anthropics Claude or the coding agent Cursor to a separate dedicated investment account. There, the agents can access the dedicated funds and place trades as directed. For example, users might instruct their agents to root out risks created by being overly concentrated in one part of the market, or monitor a basket of promising semiconductor stocks. But investors instructions can be far more detailed. In a prompt offered by Fedapuria, a customer might ask an AI agent to invest $100 based on the following parameters Sift through startup funding, deal activity and private company valuations to identify places where private market investors are putting their before they have been discovered by the public markets. For now, only stock trades are on the table. Options, crypto and event contract capabilities will come later, executives said. Investors get a push notification every time the agent makes a trade and can see a real time activity feed in the Robinhood app. They can also disconnect the agent at any time. Robinhood will also let customers connect an AI agent to a virtual version of their gold credit cards, allowing the agents to hunt for low prices, monitor availability and even make purchases that follow users instructions. The tool could monitor hard to get restaurant reservations, book a flight or snag tickets to a Broadway show on a specific day under a certain price limit, said Deepak Rao, the vice president and general manager of Robinhood Money. Agents are restricted to the virtual card only, Rao said, and can't access a customer's primary credit card number or any other account information. Customers can also set spending limits for the agent or ask to approve every purchase the agent makes on their Gold cards. We have another trillion dollar market cap company here in the U.S. micron hit the $1 trillion market value limit for the first time on May 26 after its stock closed up 19.29%, rising from $700 billion earlier in May. Driven by high memory chip demand. It is reportedly the fastest stock ever to hit the 1 trillion market cap mark. Quoting c the stock surge came as UBS tripled its price target on the stock from $535 to $1,625 a share, citing long term agreement opportunities with potentially fixed pricing. We believe the market will start to put a more normal multiple on the stock and MU will continue to re rate higher as more details emerge about the structural changes AI has driven to the entire memory complex, the firm wrote. The new price target suggests shares could more than double from Friday's close. Micron is among a fresh crop of chip makers benefiting from stage of the AI race. Investors are snapping up stocks tied to central processing units and memory needed to run and process agentic workloads in a battleground once dominated by Nvidia. Explosive demand for AI has led to a global memory shortage that chip makers like Micron are struggling to fill. That's allowed Micron and peers Skhenix and Samsung to hike prices. Micron stock has more than tripled year to date. Just a few weeks ago, Micron surpassed a $700 billion market valuation and soared into the ranks of most valuable US Tech firms. Intel, after missing on the early AI rally, is up more than six fold and trading near all time highs. The American chipmaker is in the middle of a major turnaround following a significant investment from the US Government last summer. Qualcomm, Advanced Micro Devices and Marvel Technology have also reached new highs with its dizzying rise. Micron is the 12th US company worth $1 trillion and the first based in Idaho, cementing the fastest run to a 13 figure valuation ever seen. The company reached the threshold only 48 days after it was first valued at $500 billion, according to Dow Jones Market Data. Nvidia, the undisputed champion of artificial intelligence chip makers, took 490 days to pass the same benchmark. Micron for years was considered just a commodity play. They make very basic, fairly simple things, said Michael Rosen, chief investment officer at Angela's Investments. Now Micron is the poster child. The stock might have yet more room to run as Micron builds capacity to accommodate the biggest supply crunch the memory industry has seen in over 40. Over in Korea, Skhenix topped $1 trillion in market value after its stock jumped 9% on May 27, becoming the third Asian company to hit that milestone. And it is up more than 1,000% in the past year. YouTube is making its AI content labels more prominent on desktop and mobile and will apply them automatically if it detects what it calls significant photorealistic AI use, quoting variety under YouTube's guidelines, creators will still be required to manually disclose when they use realistic AI. But starting this week, it will also roll out a new internal system to help identify AI generated content. If a creator doesn't specify whether or not they used AI, but our systems detect significant photorealistic AI use, we will now automatically apply a label, YouTube said. YouTube creators who believe their content was incorrectly flagged as AI generated can modify the disclosure status using the YouTube Studio tool. However, according to YouTube, the AI labels will remain perfect permanent in some cases, including for content created using YouTube's own AI tools such as Veo or DreamScreen, and for content that contains C2PA metadata based on standards from the Coalition for Content Provenance and Authenticity that indicates it was fully AI generated. In addition, YouTube is moving the disclosure label for photorealistic and meaningfully AI altered or AI generated content to a more prominent position. Until now, YouTube labeled AI content in a video's expanded description, but going forward for long form videos, the AI label will now appear directly below the video player and above the description. For YouTube shorts, the labels will appear as an overlay on the video itself. The goal here is context at a glance. If it looks real but was made with AI, viewers will know immediately, Rene Ritchie, YouTube head of editorial and creator liaison, said in a video about the changes. He added that the AI labels alone do not affect how videos are recommended or whether they can earn money. This is purely about giving viewers the right information at the right time. Meanwhile, for content that YouTube determines is unrealistic, animated or slightly altered but not fully AI generated, disclosures will continue to appear in the Expanded Description section. The Updates come after YouTube earlier this month expanded its likeness detection program to all creators 18 and older. That's designed to help users detect and manage how AI is used to depict you on YouTube. For creators who enroll in the program, YouTube's system will identify videos that may be altered or synthetic uses of their facial likeness. They can then request removal of unauthorized content that you uses your likeness directly in YouTube Studio. Roku has launched its first major home screen overhaul in over a decade, including a large marquee ad spot to tout apps or shows in a bid to drive more engagement, quoting the Hollywood Reporter. The new home screen will lean into personalization while making sure that it remains the simplicity that it has always been known for, said Roku VP of Viewer Product Preston Smalley, speaking in a briefing in New York Wednesday morning. The big change is your most used apps will now be featured more prominently, reducing the need to hunt for specific streaming services. There will also be a Top Picks for you section that recommends apps and programming that Roku thinks you will enjoy, and a large marquee ad spot that could tout apps or shows. There will also be genre based destinations based on your usage habits and around your subscriptions. Roku City will also get its own tile for quick access to the screensaver, which will now get an interactive overhaul. And Roku will launch a curated you'd daily scoop featuring shows and cultural trends. There will also be changes to the bread and butter functionality like search menus and shortcuts. The new home screen platform will also be able to adapt to how households use Roku's this is one of the last remaining shared devices the TV and the house, and we know that there's multiple people living in homes, smiley says. So a Roku used in a kid's playroom may have a different home screen than a Roku in an adult's bedroom, and so on. The overhaul matters because Roku is the gateway to streaming video for more than 100 million households, so any tweak to the user interface could have a significant impact. And the more prominent ad placement could help drive revenue, though Roku executives were careful to note that they don't want it to detract from the overall experience. End quote.
- [10:33]
C
Ready to soundtrack your summer with Red Bull Summer All Day Play? You choose a playlist that fits your summer vibe the best. Are you a festival fanatic? A deep end dj a road dog or a trail mixer. Just add a song to your chosen playlist and put your summer on track. Red Bull Summer All Day Play Red Bull gives you wings. Visit red bull.com brightsummer ahead to learn more. See you this summer.
- [11:00]
B
Sure, AI is everywhere, but that doesn't mean enterprise value is a given. In a recent survey, PwC found the amount of CEOs who reported re gains or cost reductions from AI is nearly equal to the amount who say they're still stuck. So what's causing the issues? PwC boiled it down to clarity. Leaders aren't clear about what's hype, what's reality, or where AI can actually create measurable impact. To help change that, PwC is offering their AI expertise and data. They explore how to tune out noise around AI and get clarity on what successful adoption looks like. Learn from the experts by heading to pwc.com US Brewai that's pwc.com US BrewAI
- [11:49]
A
so good, so good, so good.
- [11:51]
C
Everything you want for summer is at Nordstrom Rack stores now and up to 60% off. Stock up and save on the brands you love like Vince Sam, edelman frame and free people. Join the NordicLub to unlock exclusive discounts. Shop new arrivals first and more. Plus buy online and pick up at your favorite Rack store for free. Great brands, great prices. That's why you rack
- [12:17]
D
the Wired newsroom is known for award winning reporting on how technology shapes our world. On WIRED's Uncanny Valley, we take that curiosity even further. Each week, journalists from Wired break down the biggest stories in tech while speaking directly with the people building challenging and reshaping the future. Is the AI boom sustainable? How do you protect your privacy in an age of constant surveillance? Uncanny Valley tackles the questions driving today's tech debates and lighting up your group chats. Listen to new episodes every Thursday wherever you get your podcasts.
- [12:48]
B
One of the interesting things going on right now is the whole back and forth of which model is the best at any one moment. Six months ago, Claude was clearly on top, but in recent weeks, GPT 5.5 has been impressing people, especially with its coding and agentic stuff. And a new benchmark suggests that GPT 5.5 might again, at least for the time being, be further ahead than people thought, quoting VentureBeat. For months, the leading AI coding benchmarks have told enterprise buyers a comforting but misleading story. The top models are all roughly the same. OpenAI's GPT5 family, Anthropic's Claude Opus and Google's Gemini Pro have clustered within a narrow band on scale AIs swe bench pro leaderboard, making it nearly impossible for engineering leaders to determine which agent will actually perform best inside their code bases. On Monday, a startup called Data Curve released a benchmark it says shatters that illusion. DeepSUI, a 113 task evaluation spanning 91 open source repositories and five programming languages, produces a dramatically wider spread among the same frontier models and crowns OpenAI's GPT 5.5 as the clear leader at 70%, 16 points ahead of its nearest competitor. On public leaderboards, top models often look relatively close in capability, wrote Data Curve co author Serena G. On X, deepsui shows where they actually diverge, reflecting the realistic experience of developers in their day to day work. The benchmark also delivers a pointed critique of the evaluation infrastructure the AI industry relies on to measure progress. Data Curve's audit found that Swabench Pro's verifiers, the automated graders that determine whether an agent solved a task, issued incorrect pass fail verdicts on roughly one third of the trials it reviewed. If that finding holds up, it has sweeping implications. Enterprise procurement teams, venture capitalists and AI lab marketing departments all lean heavily on benchmark scores to make multi billion dollar decisions. A 32% error rate in the most widely cited coding benchmark suggests the industry may have been navigating by a broken compass to understand what Data Curve is claiming. It helps to understand how coding benchmarks work and how they can go wrong. The dominant paradigm pioneered by the Sweebench family, maintained by scale AI and academic researchers, constructs tasks by mining real GitHub commits. The process extracts a bug fix or feature addition from a repository's history, rolls the code back to the prefix state, and then asks an AI agent to reproduce the change. The original commits test suite serves as the verifier. If the agent's patch makes the same test pass, it gets credit. This approach has an elegant simplicity, but Data Curve argues it introduces three systemic weaknesses. First, there's contamination because tasks are drawn from public GitHub history. The problem statement, the discussion, and often the exact solution are already present in the training data of frontier models. The SWE bench Family scrapes existing GitHub issues and PRS, which creates two problems. Memorization models have already seen the solution and triviality. Most tasks are small, g wrote. Second, there's scope. SweetBench Pro tasks require on average just 120 lines of code added across five files. Deep Sweeze reference solutions average 668 lines added across seven files, roughly 5.5 times more code. Yet Deep Sui's prompts are actually shorter, averaging 2,158 characters versus Sui Bench Pro's 4,614. In other words, DeepSui gives the agent less instruction but expects far more output, which more closely mirrors how a human developer might actually delegate work to an AI assistant. Third, and most damaging Verifier Reliability data curve drew 30 tasks at random from both Deepsui and Swebench Pro, ran three rollouts across 10 frontier model configurations, and then deployed an LLM based judge to independently assess whether each agent's patch actually solved the problem. Suitebench Pro's verifiers accepted wrong implications 8.5% of the time and rejected correct implementations 24% of the time. Deep Sweeze verifiers registered just point and 1.1%, respectively. The false negative problem is especially insidious because it punishes creative solutions. In one documented case, the gold standard pull request for a SU Bench Pro task refactored a private helper function, an agent that correctly solved the task by inlining the same logic. A perfectly valid engineering choice failed because the test suite tried to import a symbol that only existed in the original author's specific implementation. Deep swee's top line results reorder the familiar hierarchy in ways that should matter to every engineering team evaluating AI coding tools on SUI Bench Pro models from OpenAI, Anthropic and Google have traded the lead within a 30 point range. Deepsui stretches that range to 70 points, GPT 5.5 leads at 70%, followed by GPT 5.4 at 56% and Claude Opus 4.7 only at 54%. From there the drop off is steep. Claude Sonnet 4.6 lands at 32%, Gemini 3.5 flash at 28%, GPT 5.4 mini and Kimi K2.6 tied at 24%, and then a long tail of models in the teens and single digits. Claude Haiku 4.5, which scores 39% on Sui Bench Pro, collapses to zero on Deepsui, suggesting that some mid tier models have been significantly overperforming on easier, potentially contaminated benchmarks. GPT 5.5 doesn't just score the highest, it does so efficiently. The Model reaches its 70% pass rate with a median cost of $5.80 per trial, a median wall clock time of 20 minutes, and a median 47,000 output tokens. GPT 5.4 emerges as perhaps the best overall value at $3.30 per trial with a 56% score Claude Opus 4.7, meanwhile, costs significantly more per run, and output tokens, wall clock duration, and dollar cost per trial all vary by an order of magnitude across agents tested, yet none of these correlate strongly with pass rate. Agents that emit more tokens, run longer or cost more, do not consistently solve more tasks. End quote. Nothing more for you today. Talk to you tomorrow.