$4STOCK has probably been one of the hottest tickers on BSC over the past few hours. I think most people already have a rough idea of what 4Stock does, so I don’t want to turn this into another docs recap What I care about more is : Why is the market pricing this narrative so aggressively, what is BNB actually trying to do with meme x stocks, and how do I view $4STOCK after its first major run? 1. Quick recap of the narrative 4Stock is new direction for Stock Memes on BNB Chain For stocks that don’t have a bStock yet, can bring them onchain first through 4Stock Users send USDC, uses that capital to purchase the actual underlying stock, then mints an equivalent amount of 4Stock based on the number of shares actually purchased Those assets can then be traded onchain, but more importantly, they can also be used as base pairs for communities to launch Stock Memes on The basic flow is : real stock => 4Stock => base liquidity => Stock Meme And once an official bStock for that underlying becomes available, the corresponding 4Stock is designed to be convertible into that bStock But that’s the product layer The $4STOCK token we’re trading is not a stock tokenized 1:1 I see it more as a meta token betting that this entire Stock Meme machine becomes a major narrative on BNB That distinction is the first thing you need to understand before playing this 2. The timing of this narrative is pretty good imo Meme x stocks isn’t exactly new If you’ve been following my posts over the past few days, you already know this wave ran hard on Robinhood Chain before attention started rotating toward SOL BSC also tried to catch the wave with $BREW, but imo after that brutal dump and all the bundle FUD, it failed to create the confidence BSC needed for this narrative But even then, I kept thinking one thing : There’s no way BNB is going to sit out a meta that’s attracting this much volume And this time, the approach looks much more structured There’s tokenized stocks, the BNC narrative, Stock Meme infrastructure and the broader Stonks season BNB Chain has also previously amplified Stock Memes on directly That obviously doesn’t mean BNB is guaranteeing the chart of any token But to me, it does show that the ecosystem actually wants to bring this rotation onto BSC, rather than this simply being a story created by a few KOLs around a random ticker 3. BNC4 is what made me pay more attention to the thesis $BNC4 is the first 4Stock and probably the most interesting proof of concept so far BNC already has its own narrative, with the market viewing it as something like a “MicroStrategy for BNB” chose BNC as the first underlying, brought it onchain as BNC4, and then allowed BNC4 to become a base asset for Stock Memes More importantly, there has been real demand said it received more than $2M in mint requests shortly after launch But that number needs to be understood correctly This is the amount of mint requests reported by not onchain proof that every dollar of that $2M+ has already been processed and minted first needs to use the USDC to purchase the underlying BNC shares, then mint BNC4 according to the actual number of shares purchased So additional BNC4 supply can continue entering the market as those requests in the queue are processed And this is where the game gets interesting 4. Arbitrage is both bullish and bearish BNC4 traded at a massive premium to the underlying BNC at one point The gap was several times the actual stock price At that point, the game becomes pretty straightforward : mint near underlying value => receive BNC4 => sell into the premium on the secondary market One notable case was 0xShawn, who minted and sold 6,666 BNC4, receiving roughly $231K USDT onchain. Media described it as an ~$230K arbitrage case while BNC4 was trading at a huge premium to BNC in after-hours trading To me, this is both bullish and bearish Bullish because it proves that the product is actually being used People are depositing capital People are minting There’s secondary liquidity There’s arbitrage And communities are starting to use these assets as base pairs for Stock Memes But it’s also bearish for the BNC4 premium itself If an asset can be minted close to its underlying value and sold onchain for several times more, the market is basically paying arbitrageurs to : mint more supply => sell the premium => pull the price back toward the underlying So if BNC4 gradually moves closer to BNC, I wouldn’t necessarily see that as the product failing If anything, that’s the arbitrage mechanism doing its job 5. But $4STOCK is a completely different game This is where I think a lot of people are getting bullish for the wrong reason BNC4 has an underlying asset. $4STOCK is not BNC4 Holding $4STOCK does not automatically give you ownership of BNC You can’t redeem it 1:1 for stock You don’t automatically receive the 1% mint fee And you don’t automatically receive 50% of Stock Meme fees either So I wouldn’t fundamental-value $4STOCK based on the amount of stock holds in custody or the amount of USDC flowing into mints What the market is actually buying with $4STOCK is much simpler : “Can become the center of Stock Meme season on BSC?” If the answer is yes, the market could potentially treat $4STOCK as an attention/index token for the entire category But that’s narrative premium Not NAV 6. So what am I actually bullish on here? I’m actually more bullish on the broader BNB Stock Meme season than on any single ticker The current flow looks pretty interesting to me: BNB narrative => BNC stock => BNC4 => 4Stock infrastructure => Stock Memes => $4STOCK speculation And attention is moving through multiple layers of that flow What I like most is that BNB now has exactly what a new meta needs : a narrative that is “real” enough to support a fundamental story, but still degen enough to continuously create new coins and volume That’s why I think model is pretty clever It isn’t trying to turn meme traders into stock investors It takes something Wall Street understands , stocks - brings it onchain, then turns it into something the trenches understand best : liquidity for launching memes So this isn’t really DeFi for stocks. It’s more like RWA becoming fuel for a new meme casino 7. But the easy money on $4STOCK is different now This is probably the most important part of my view on the price action Someone who entered extremely early has a completely different risk/reward profile from someone buying after the narrative has already gone viral At this point, the market is no longer pricing an experiment nobody knows about It’s starting to price in the assumption that could win the Stock Meme season on BSC So the next leg of upside needs execution I want to see the 2nd, 3rd and 4th 4Stock I want to see Stock Memes launched from those pairs actually produce runners I want to see mint demand continue growing instead of stopping at the initial $2M+ figure And most importantly, I want to see the BNB ecosystem continue amplifying this narrative If those things happen, $4STOCK has a reason to keep carrying a premium as a meta token If they don’t, attention can rotate into new tickers very quickly Especially when $4STOCK itself hasn’t shown me any direct product cash flow flowing back to token holders 8. So what am I watching from here? I’m not going to take the BNC stock price on Nasdaq and try to calculate some “fair value” for $4STOCK because there is no direct NAV relationship between the two I’m watching four things : Stock Meme volume - is it actually growing, or is all the volume just rotating around $4STOCK? Mint demand - does demand for BNC4 and future 4Stocks continue? New underlyings - does keep bringing new stocks onchain, or does the entire narrative stop at BNC4? And finally, BNB attention - does the ecosystem keep pushing this, or does attention rotate into another meta? If all four continue expanding, the thesis is being validated If $4STOCK keeps pumping while the activity underneath it stays flat, I’ll become much more cautious A meme token trading ahead of its fundamentals is completely normal What isn’t normal is looking at that market cap and assuming there is an equivalent amount of real assets sitting behind it 9. Final thoughts After what happened with $BREW, I didn’t think BNB would let meme x stocks end there Robinhood Chain opened the wave SOL started following And now + the BNB ecosystem seem genuinely interested in bringing that rotation onto BSC What makes 4Stock interesting to me isn’t simply the chart The underlying product is real Mint demand is real The arbitrage is real And the mechanism of using onchain stocks as base assets for Stock Memes has started working Meanwhile, $4STOCK is the ticker the market is using to speculate on whether all of this becomes an actual season So I’m pretty bullish on the narrative, but I’m not going to call $4STOCK an RWA or pretend its market cap is backed by an equivalent amount of stocks If keeps adding new stocks, creates more runners and BNB continues pushing the narrative, I think $4STOCK is positioned pretty well to become one of the main meta tokens representing the entire wave But if everything stops at BNC4 and a few days of initial hype, that premium can disappear very quickly too In one sentence : BNB is experimenting with turning stocks into gas for memes. The experiment is real. $4STOCK is still the bet on whether that experiment succeeds 0xd270D4e1EC6e6E0d28C0ecB8BE966EC75997FFfF

Tons of interesting things in the Kimi K3 tech report — here are five algorithm-side techniques that I think either I've never seen before or simply deserve more attention than they're getting. 1/ They open-sourced the model but kept the speculative decoding draft model, which could be a real serving advantage of their own. Quick background for those less familiar — speculative decoding pairs a large target model with a small draft model. The draft cheaply proposes several tokens ahead, and the target verifies them all in one forward pass. The speedup is determined by the acceptance rate, i.e., how often the target agrees with the draft's proposals. K3 is pre-trained with an MTP (multi-token prediction) layer, DeepSeek-V3 style — an extra layer on top of the backbone that predicts one token further into the future than the main next-token head. Structurally this layer is an exact copy of a regular backbone block, so you can think of it as a 94th layer that has been trained on the full pre-training corpus from day one, just with a shifted prediction target. After post-training, they freeze the target and fine-tune this MTP layer into an EAGLE-3-style draft. The EAGLE family of methods makes the draft a single decoder layer that reads the target model's internal hidden features rather than only the generated token sequence — conditioning on the target's features is what lets a one-layer draft stay accurate. The fine-tuning is then set up to match inference exactly. At inference, the draft proposes multiple tokens in a row, so from the second token onward it is building on its own unverified guesses rather than anything the target has confirmed. They replicate this condition during training by unrolling the draft for 7 steps — the first step uses the target model's features, and every step after that consumes the draft's own outputs from earlier steps. Two more design choices worth knowing. The draft reads low/mid/high-level target features (outputs of the 1st, 4th, and final AttnRes blocks), concatenated and passed through a fusion matrix initialized as [0 0 I] — zero weights on the low and mid features, identity on the high-level one. At initialization the draft therefore sees exactly the high-level feature the MTP layer was pre-trained on, and it gradually learns to mix in the other two during fine-tuning. And instead of the usual KL surrogate, they directly minimize the negative log of the acceptance rate itself (the sum of min(p, q) over the vocabulary, where p and q are the target and draft distributions), since minimizing KL does not guarantee maximizing acceptance for a capacity-limited draft. Everything is trained under the same MXFP4/MXFP8 QAT as their serving stack. The release itself is asymmetric. The full target weights are on HuggingFace, but the draft — and as far as I can tell, the MTP layer it was fine-tuned from — is not. That MTP layer was trained jointly with the backbone on the full pre-training corpus, which no external party has access to. So first-party serving stays faster and cheaper on the exact same open weights. This is the smartest business decision I've seen recently from open-weight model companies. 2/ Sync RL with partial rollouts. Sync RL waits for every rollout in the batch to finish before updating, and since rollout lengths vary wildly, compute is wasted by waiting on all rollouts to complete. Async RL decouples actors from the learner, which is much more efficient, but actor weights go stale, and a long rollout can land several learner steps behind the current policy. K3 runs a middle ground, where generation pauses as soon as a fraction of the trajectories completes, and optimization proceeds immediately, like async. However, unfinished trajectories get paused, enqueued, and resumed at the start of the next iteration under the freshly updated policy. In other words, a single 1M-token trajectory can literally be a relay across several different policy versions. Essentially, they trade model staleness for data staleness — off-policy prefixes inside otherwise on-policy trajectories — and mitigate it with a per-token regularization that constrains each update to a localized neighborhood of the current policy. It's an intellectually pleasing trade. What makes this viable at 1M context is the environment side, since pausing the model's rollout is easy; but pausing a live agentic sandbox mid-trajectory is not. Their microVM runtime checkpoints an environment in 133ms and resumes it in 49ms, and a paused sandbox consumes zero CPU and memory. 3/ Multi-teacher on-policy distillation as the merge step. After RL, they have nine expert models — three domains (general / agents / coding) crossed with three reasoning-effort levels (low / high / max) — consolidated into one unified model through multi-teacher OPD. The idea itself is not new; the report cites the same lineage as Thinking Machines' OPD post, MiMo-V2-Flash, and DeepSeek-V4. Three details stand out though. First, this is distillation with zero compression. Teachers and student are the same 2.8T architecture, and OPD is purely the mechanism that folds nine RL policies into one model, not a way to shrink a big teacher into a small student. Second, the OPD signal is implemented as a per-token RL reward — the clipped log-ratio between the teacher's and the student's probability of each generated token. Distillation is therefore not a separate pipeline; it is literally the same RL trainer running with a different reward. The student generates its own on-policy rollouts, the teacher scores every token along the way, and everything above carries over for free — partial rollouts, pausable sandboxes, the per-token regularization — which is what makes it feasible to distill even million-token agentic trajectories. Third, a negative result. At each step the student samples one token from its own distribution and it is all the OPD reward looks at, requiring just a single number per step, the teacher's log-prob of that token. They experimented with finer-grained top-k objectives that match more of the teacher's distribution over candidate tokens at each step, and saw no advantage in either convergence speed or final performance. So they decided that no full logits were needed. 4/ The RL harness is randomized. They represent an unified agent harness with shared tool interfaces, system prompts, context management strategies, skills, memories, subagents, and can instantiate Kimi Code, Claude Code, Codex, OpenClaw, Hermes, or entirely new harnesses from the same abstraction. During RL, harness configurations are dynamically reshuffled across task groups so the model never overfits to any single tool schema or interaction protocol. The implication is that harness generalization is a trained property, not an emergent one. If you have ever evaluated open models across different agent scaffolds and wondered why some transfer well and some fall apart, this is probably a big part of the answer. It also fits Kimi's position as an open-weight company. A closed lab ships the model and the harness together and controls the whole stack; an open model gets dropped into whatever scaffold people already use — Claude Code, Codex, OpenClaw, some custom internal agent, so harness robustness is even more importnat for open-weights models. Interestingly, their own in-house coding bench even reports K3 scoring slightly higher under Claude Code than under their own Kimi Code. 5/ NoPE on every global attention layer. All 24 Gated MLA layers in K3 use no positional encoding at all. Positional and recency information is carried entirely by the KDA layers' gating and decay (the backbone runs 3 KDA per 1 MLA), while the MLA layers do pure content-based global lookup. The payoff shows up at context extension. K3 grows from 8K to 64K during pre-training and from 256K to 1M during cooldown with zero positional-encoding modification — no RoPE base retuning, no YaRN. Hybrid linear attention is usually pitched as the efficiency component of these architectures; here the linear layers are also doing the entire job of the position encoding. Honestly, this only scratches the surface. The infra sections (MoonEP, quantile balancing, KDA-aware prefix caching) each deserve a post of their own. Full report is definitely worth the read.

Once again, The Failing New York Times has attempted to grossly mischaracterize what should be hailed as the restoration of Beauty and Grandeur to our Nation’s Capital as something else completely. Lafayette Park, across the street from the White House, had been left in disarray after decades of neglect and lack of maintenance. The fountains didn’t work, the grass and trees had died, the walking paths and benches were almost unusable. Also, because of security threats caused by rioters and protestors ripping up the stone pavers, and throwing them at our brave men and women from the Secret Service and Law Enforcement, my Administration decided to rebuild and beautify the Park. I made a multimillion contribution to the effort, and got others to do the same, but was not in charge of handing out the contract. That was done by the National Parks Service, and they gave it to the largest and most respected Construction Firm, for many years, in D.C., Clark Construction — A greatly respected firm, by far, the biggest in D.C. Not only did we add brand new fountains, but we completely beautified the Park, adding a new irrigation system, sod, trees, park benches, and more. It’s turning out magnificent, under budget, and way ahead of schedule! We want to have it complete by July 4th. Once again, people will come to the front door of the White House by walking through a magnificent Park befitting these Hollowed Grounds, not a dead, dirty, rusting, and very dangerous place like it was before we got involved. In fact, D.C. itself is now a safe and bustling place again. The Crime Numbers are the best they’ve had in Recorded History. The New York Times should congratulate us, instead of trying to make us look bad. We look forward to continuing to MAKE WASHINGTON, D.C. GREAT AGAIN, and want to thank Clark Construction, the Department of Interior, and the National Parks Service on A JOB WELL DONE! President DONALD J. TRUMP ( Donald J. Trump - TS: Apr 25 2026, 4:18 PM ET )