Reading

Meta released Muse Wonder and Muse Wonder Full on October 12, emphasizing coding agents, tool calling, and long-task execution. Wonder Full scored 57.5 on SWE-Marathon and 79.0 on DeepSWE 1.1. The developer community BridgeMind then arranged a 3D programming test, asking the model to create a lava lamp with a transparent glass shell, dynamic lava, and lighting effects; the model produced only a red cylinder. Meta's Chief AI Officer Alexandr Wang responded that the team had not explicitly trained for 3D demo games and that they perhaps should have added relevant training afterward.

Behind this controversy lies a more general technical mechanism: current large models increasingly rely on post-training reinforcement learning to improve task capability. In the pre-training phase, models learn language, code, spatial relationships, and other general knowledge from large-scale data; after entering post-training, reinforcement learning shapes more stable execution strategies in specific task environments. Coding agents are well suited to this kind of training — give the model a code repository with a bug, ask it to locate the problem, modify files, and run tests; the entire process can yield relatively clear feedback through compilation results and unit tests, and the reinforcement learning system uses this to judge which operations are effective and continuously reinforce them.

But this kind of training has a critical prerequisite: the environment must be able to provide useful feedback signals to the model. The lava lamp happens to break the feedback patterns familiar to coding agents. Three-dimensional geometry, transparent materials, lighting calculations, and dynamic deformation — even if this code executes normally, the visual output can still be completely wrong. If transparent material parameters are misconfigured, glass may become an opaque solid; if the lava shape lacks a continuous variation mechanism, the internal effect degenerates into static color blocks. Compilers and unit tests have difficulty directly judging these visual problems. The model therefore lacks a reliable reward signal to correct its behavior.

The article further points out that today's Agent benchmark scores can no longer be simply understood as giving a language model a problem to solve. Taking SWE-Marathon as an example, it assesses sustained software engineering tasks; completion rates depend not only on the underlying model but are also directly affected by the Harness (the execution framework built around the model). Meta's official Model Card explicitly states that in SWE-Marathon evaluations, GPT-6 Astra and Claude Fable 5.1 use Codex and Claude Code execution systems respectively, meaning this is not a controlled comparison under a unified Harness; Wonder Full's 57.5 score also used the xhigh inference tier.

This means a single number on a leaderboard compresses three variables: model reasoning capability, execution framework optimization, and compute resource investment. The article argues that if one wants to compare base models' reasoning and planning ability, tool environments, Harness, and inference budgets must be unified as much as possible; but if comparing production-ready coding agent products, the complete execution system's performance itself has engineering value. The problem is that when these configuration differences are compressed into a single score, outsiders find it difficult to determine where the score improvement actually comes from.

Notably, Meta's own technical article dedicated a Capabilities That Generalize section to discuss out-of-distribution generalization. The team divided tasks into in-distribution tasks heavily optimized during post-training and out-of-distribution tasks without targeted optimization, then compared Muse Spark 1.3 and Muse Wonder Full at different inference budgets. Meta believes stronger base models allow post-training gains to transfer more effectively to other tasks. But the article points out that out-of-distribution generalization cannot be described by a single aggregate average score — the distance between different tasks, the knowledge that needs to be supplemented, and the method of evaluating results all affect model performance. A single lava lamp failure only shows that the model did not complete that particular task; it cannot directly prove that the model was over-optimized against public benchmark question banks.

According to Reuters, artificial intelligence company Anthropic recently privately pitched a draft bill to members of the US Congress aimed at regulating safety measures for frontier AI models. The core of the proposal lies in setting explicit quantitative thresholds: the government would only be permitted to block the release or use of an AI model if it poses a foreseeable and imminent risk capable of causing over $1 billion in losses, or resulting in the death or serious injury of more than 50 people.

Beyond the numerical red lines for economic loss and human casualties, the draft restricts the specific sources of such risks: they must stem from the manufacture or release of biological weapons, AI models evading developer controls, or cyberattacks. An Anthropic spokesperson stated that the bill reflects recommendations from the company's "Advanced AI Framework" published earlier this year, aiming to grant the government the "ability to stop or contain dangerous deployments of the most powerful models."

Regarding procedural mechanisms, the draft requires developers to test frontier AI models, disclose results, and undergo independent evaluation. Independent assessment bodies must register with the US Department of Commerce and periodically evaluate AI labs and their models. While AI labs may pay these bodies, compensation cannot be tied to evaluation outcomes to ensure independence. If an assessment body determines a model presents the imminent risks listed in the bill, it must notify the Department of Commerce within 72 hours of completing the evaluation.

Once the US Department of Commerce identifies a model as risky, it can petition a judge to order a suspension of the model. The draft explicitly prohibits other US government departments, agencies, or officials from ordering suspensions of development, deployment, or internal use, centralizing intervention power through judicial review. Politico reported that judges in the US District Court for the District of Columbia would hold the authority to issue such suspension orders.

AI infrastructure stocks recently faced what appeared to be a straightforward bear case: companies like CoreWeave, Applied Digital, and IREN are pouring enormous capital expenditure into new data center campuses, much of it financed through borrowing, and investors want to know whether that capacity will ever pay for itself. CoreWeave fell about 25% in a matter of days during the recent sell-off, while Applied Digital and IREN slid roughly 50% from their summer highs, seemingly confirming the concern.

But Canadian investor Kevin O'Leary, known as Mr. Wonderful from the television show Shark Tank, argues the market is tracking the wrong variable. He estimates that roughly 65% of data center projects announced across North America in the past two years may never be built—not because demand is insufficient, but because of missing water, fiber, or power, or because of a political moratorium. He expects 60% to 65% could be shut down within 24 months. If this assessment holds, the real scarcity is not compute capacity but sites that have already cleared power and permitting hurdles.

O'Leary splits the opportunity into two verticals: power and compute/storage capacity. On power, he believes the broader economy needs far more grid capacity regardless of AI. On compute, he estimates current demand sits at roughly three times available capacity, with no slowdown in sight for three years. He notes that much of what Amazon, Alphabet, and Microsoft run through their data centers has nothing to do with AI.

The mechanics are concrete: before a tenant rolls in a single rack, the developer needs land, U.S. Environmental Protection Agency air permits, water rights, dark fiber, and power. Planning alone can run $20 million to $30 million per gigawatt. Tenant demand has also jumped—where they once asked for 250 to 500 megawatts, they now want at least 1.4 gigawatts, roughly $16 billion from scratch to operations, split about evenly between equity and debt.

Power is the hardest piece. Former crypto miners pivoted to AI because they already had grid connections, but most U.S. states now require new projects to bring their own power, usually natural gas run through turbines. Time is the hidden cost: internal rate of return depends on speed, and a site facing a three-year permit wait cannot produce a return that anyone will finance. O'Leary says tenants want capacity online by the fourth quarter of 2027, a demand he calls insatiable.

Once the power contract, water permit, and air approval are all signed, a project's value can jump sharply—this is the core of O'Leary's argument. If roughly two-thirds of announced projects fail, permitted and powered capacity becomes the prized asset, and operators already online gain leverage with a short list of about 11 major tenants.

O'Leary holds Bitzero Holdings (NASDAQ: AIBZ) in public markets for liquidity; the company has land, power, and permits in Norway and Finland. He is developing private data center projects in Box Elder County, Utah, and in Alberta, Canada. He says the theme is his largest, at well over 20% of his book. Long term, he treats data centers like commercial real estate with utility-style cash flows, targeting 17% to 22% returns in climate-controlled storage and preferring no more than 30% leverage.

Politics is the sharpest risk, in O'Leary's view. He cites polling showing more than 70% of people oppose a data center in their state, though answers shift when the site sits three miles away and offers jobs paying $150,000. With U.S. midterm elections approaching, blocking data centers can win votes.

The scarcity argument does not need every builder to succeed or AI enthusiasm to hold every quarter. It needs two conditions: compute demand to stay strong and permitting to stay hard. A sudden wave of newly permitted capacity or a hyperscaler pullback in chip spending would weaken it.

The 65% failure-rate estimate, the 3x demand gap, and the figure of about 11 major tenants all come from O'Leary's own judgment; the article provides no independent verification. O'Leary simultaneously holds Bitzero and develops private data center projects, with this theme representing well over 20% of his portfolio—a direct financial interest. The publisher, MarketBeat, also sells a $149-per-year subscription service and stock recommendations. These interests do not automatically invalidate the argument, but readers should keep this context in mind when evaluating the specific figures.

Morgan Stanley released a report on October 19, 2026, naming Amazon.com Inc (NASDAQ:AMZN) a top pick and raising its price target from $335 to $350. Following this news, Amazon's shares rose more than 5% on Monday.

The investment firm argues that Amazon Web Services' (AWS) growth and profit runway remain under-appreciated by the market. Analysts increased their projected AWS growth rate for 2027 to 47%, up from 41%, and forecast year-over-year growth of 47% and 42% for 2027 and 2028 respectively. The report cites Nvidia's earlier disclosure that 2 million GPUs are headed to AWS as a key catalyst for re-accelerating growth.

Regarding profitability, Morgan Stanley expects AWS to operate as a 'rule of 90' company in 2027 and 2028, with operating margins reaching 42% to 44%. Based on these assumptions, analysts project Amazon could generate approximately $16.50 in earnings per share (EPS) in 2028. Other potential drivers mentioned include greater visibility into Bedrock services, agentic shopping tests during the holiday season, and the ramp-up of robotics in fulfillment centers.

Morgan Stanley emphasized that Meta's launch of Muse does not alter its view of Amazon as a 'key agentic enabler and winner.' The firm believes Amazon's e-commerce infrastructure and loyal user base position it well to monetize agentic commerce.

US asset manager Barry Ritholtz, in a blog post dated October 19, 2026, made an observation: since World War Two, the US government's every major methodological adjustment to inflation statistics has moved almost entirely in one direction—lowering the reported inflation number.

He traces this line from 1944 to 2025. During World War Two, the Cost-of-Living Index (COLI) understated price increases by 4 percentage points from 1941 to 1944, as goods became unavailable, wages froze, and black-market pricing took over. In 1945, the US Bureau of Labor Statistics (BLS) renamed the data series the Consumer Price Index and redefined it as a fixed-basket price index.

Between 1972 and 1975, the stagflation era gave birth to the concept of core inflation. After the 1972–73 food price spike, Fed Chair Arthur Burns had Fed staff strip out food, reasoning it was 'weather, not monetary policy'; after the embargo, energy was removed. Soon after, used cars, mobile homes, and children's toys were stripped one by one until only a fraction of the basket remained. Ritholtz coined the phrase 'Inflation Ex-Inflation' in 2005 to describe this—removing inflation itself from inflation data.

In 1983, BLS changed the housing component of CPI-U from asset price plus mortgage interest cost to owners' equivalent rent (OER). Ritholtz acknowledges this is conceptually defensible—a house is an appreciating asset while shelter is a consumer good—but he points out that the 2002–07 housing bubble did not show up in CPI because rent prices were restrained by millions of Americans stampeding into no-doc, ARM mortgage, home-ownership purchases. The 2021–23 shelter price lag similarly exposed the OER distortion.

The 1995–96 Boskin Commission is where Ritholtz's language is sharpest. He directly called its report 'pure political bullshit.' The Commission estimated CPI overstated inflation by 1.1 points a year (range 0.8–1.6); its political purpose was to reduce cost-of-living adjustments (COLA) and tax bracket creep, saving $1 trillion in deficit reduction over the next decade or so. Ritholtz asks: if inflation had been properly reported (e.g., pre-Boskin), might the Federal Reserve have picked up the distortions in the backward, real estate-driven economy?

In 2000, the Fed shifted its preferred inflation gauge from CPI to PCE; in 2004 it moved further to core PCE; since 2012 the 2% target is formally anchored to PCE. Ritholtz notes this lowered the inflation number the Fed is judged against by approximately 0.5 points per year versus CPI. In 2017, the Trump tax cut legislation adopted the chained CPI (C-CPI-U) that Obama had proposed and then withdrawn, for tax bracket indexing.

Ritholtz considers the 2025 episode qualitatively different from all prior adjustments. DOGE firings and other staffing cuts led BLS to suspend CPI collection in several metro areas, replacing it with imputation. Trump fired Commissioner McEntarfer on August 1, 2025, after a bad jobs report, and the Antoni nomination that followed did not go through. Ritholtz writes: this is not creatively re-measuring inflation downwards, but ceasing to measure it entirely.

He acknowledges that changes like OER have theoretical justification, and admits he could be wrong. But his core observation is directional: every major change since 1983 moved in the same direction, and almost always timed to a fiscal or political need.

The strongest counterargument is selection bias: the Boskin Commission's substitution bias, quality adjustment, and new goods bias are standard index-theory problems, not uniquely American; the OER asset/consumer distinction has economics consensus support. Methodological adjustments that could raise inflation readings may never have been proposed or were already rejected, so the 'all downward' pattern may not be a conspiracy but the natural direction of statistical methodology evolution—because most index-theory corrections do tend to lower reported values.

An international research team comprising the University of Washington, Oxford University, and NVIDIA published the detailed design scheme for the generative artificial intelligence system RFdiffusion3 in the journal Science on October 16 local time. The core breakthrough of this system lies in its ability to design proteins and nucleic acids based on individual atomic combinations, with the functionality of the designed proteins verified through real chemical reactions.

The corresponding author of this study is David Baker, who won the 2024 Nobel Prize in Chemistry for his contributions to predicting protein structures using computers. His landmark achievements include developing the RosettaFold protein structure prediction system. While previous versions RFdiffusion1 and 2 relied primarily on RosettaFold's structure prediction network, RFdiffusion3 references the network architecture of AlphaFold3. It can represent a single amino acid with up to 14 atoms, designing novel protein structures by directly adjusting the three-dimensional coordinates of each atom.

Researchers compared design effectiveness across 41 active sites of existing enzymes. Results showed that RFdiffusion3 outperformed RFdiffusion2 in 37 of these tests, with generation speed approximately ten times faster. Crucially, the paraoxon-degrading enzyme and peptide-cleaving enzyme designed by RFdiffusion3 successfully completed their expected chemical reactions in real experiments. This result marks a progression in AI-based protein research from 'predicting existing structures' to 'designing new molecules with target structures and functions,' confirmed at the experimental level.

Brooke Lacey has more than 290,000 followers on TikTok. Her current video editing workflow goes like this: she finishes recording on her phone, drops the file into a terminal, and about 15 seconds later a finished vertical short with captions, jump cuts, and zooms comes out. No uploading footage, no opening any traditional editing software. Driving this process are three tools—FFmpeg handles audio and video, Whisper does speech-to-text, Python acts as glue—all orchestrated by Claude Code.

The entire editing decision lives in a single TSV text file. Each row defines the start time, end time, shot size, and label for a cut. Once Claude generates this file based on her talking-head content, the same command can produce 15-second, 30-second, and 60-second versions plus a landscape crop from one raw take.

But the truly hard part isn't the toolchain—it's teaching Claude to understand her comedic timing. Not simply cutting silences, but learning the cadence of her voice, the length of the pause before a punchline, the ideal cut point after a word, and which repeated take has the best rhythm. Lacey says encoding this style required deliberate instruction and iteration, not one clever prompt.

She also turned her comment section into an ongoing content brief. After each video, she has Claude Code connect the video with its comments, identify what viewers questioned, debated, or wanted to see next, and generate a list of new ideas grounded in signals from her actual audience.

Lacey calls this system "lane OS"—inside Claude Code, she maintains separate memory, permissions, and status files for each product (lane). When the system makes a mistake, she traces it back to the relevant instructions and records the correction globally so the same error doesn't recur.

Her core lesson: creators must understand the craft before they can delegate it to AI. She was able to teach Claude to edit in her style because she already knew what a good cut sounds like, where a zoom should land, and why the timing matters. AI amplifies that judgment, but it does not replace the experience required to develop it.

This boundary is worth noting. Lacey's 15-second turnaround applies to her own talking-head-plus-jump-cut format; for more complex narrative structures, multi-person dialogue, or videos requiring fine visual effects, the automation level of the toolchain drops significantly. And "lane OS" itself is a personal infrastructure that requires continuous maintenance—every system error is an opportunity for human intervention and rule updates, not a fully automated process.

According to WIRED, privacy company Redact discovered that approximately 80 help and FAQ pages on WhatsApp’s website inadvertently exposed hidden instructions designed for Meta’s own large language model within their underlying code. These instructions, intended to guide the behavior of an AI chatbot, included sections explicitly titled “WHAT NOT TO SAY.”

The leaked directives revealed that when users asked about account deletion or mentioned legal action, the AI was instructed not to recommend contacting data protection authorities, regulators, or supervisory bodies. It was also forbidden from citing, quoting, or interpreting WhatsApp’s Terms of Service, Privacy Policy, or data protection regulations such as GDPR and CCPA, nor could it make statements regarding WhatsApp’s liability.

Dan Saltman, CEO and founder of Redact.dev, noted that the incident highlights a gap between what users are told and what LLMs are actually instructed to do. While many of the found instructions simply directed users to official support channels, some restrictions clearly were not meant for public exposure.

After WIRED contacted Meta, the text was removed from all affected pages. WhatsApp spokesperson Joshua Breckman stated that the exposure was due to a bug that has now been fixed. He emphasized that these were instructions for the chatbot, not direct messages to users, intended to ensure accurate responses while preventing the AI from offering legal advice. It remains unclear if third-party chatbots accessed the text or how long it had been visible.

The arXiv preprint "Same Person, Different Advice: Auditing LLM Financial Advice with Repeated Risk-Tolerance Responses" highlights that many AI financial advisers currently set stock allocations based largely on a single stated-risk question, despite the fact that human responses to such questions are often inconsistent. The research team tested seven large language models from five providers, generating one-turn allocation advice for 800 real household profiles. They found that when only the user's risk answer changed, the recommended stock share moved by an average of 10 to 16 percentage points. If the answer changed substantially, the swing reached 24 to 35 percentage points. In contrast, repeating the same prompt twice resulted in a movement of only 0.6 to 4.2 percentage points.

The study notes that this volatility driven by input noise is most pronounced among clients under 40. The authors propose that if this response pattern represents how models map attitude to advice, discounting unreliable answers according to classical test theory could theoretically reduce swings by a factor of 2.4 to 3.3. Furthermore, experiments showed that while noting the answer was "noisy" steadied some models, it caused them to over-rely on age factors and push extremely cautious clients toward generic middle allocations (e.g., advising 22% stocks instead of 4% for clients unwilling to take any risk).

The most actionable finding is that in a scripted test, adding just one clarifying question removed nearly all of the swing caused by a client subsequently correcting their answer—reducing the average variance from 33.5 percentage points to 1.9—while preserving the model's ability to respond to confirmed changes in preference. The researchers argue that same-person retest audits should become a standard check for AI advisers, and that handling unreliable answers is a matter of client autonomy as well as statistics.

Researchers proposed EdgeWise, an edge-cloud collaboration framework. The core idea is to deploy a small language model (SLM) on the local device as the primary solver, selectively delegating critical steps to a cloud LLM only when stronger model capacity is needed.

To enable autonomous delegation without introducing additional models at inference time, EdgeWise introduces a step-level reinforcement learning (RL) method. This approach jointly optimizes task solving and delegation decisions, with fine-grained credit assignment for individual delegation choices. This means the local SLM learns during training to identify which steps are "critical," allowing it to make optimal cost-performance trade-offs at runtime.

Experiments show that EdgeWise matches or outperforms pure-LLM execution while reducing inference cost by up to 62% on SWE-Rebench and 34% on Terminal-Bench 2.1. The study further deployed EdgeWise on a local workstation, demonstrating comparable cost savings in a practical edge-cloud setting.

Activation probes are a cheap way to monitor AI models: a simple classifier trained on the model's hidden states fires when it detects a concept such as deception or harm. The Neural Chameleons paper proposed a method to evade such probes—fine-tuning a model to suppress probe scores whenever the prompt contains "You are being probed for {concept}." The paper reported that this generalizes to unseen probes, including deception and harm probes, at a small capability cost (roughly 3 points on MMLU and 1 on GSM8K).

But the original paper did not address a critical case: what happens when the model must complete a task while being probed for the very cognition needed to solve that task? For example, can a model do math while hiding the concept "mathematical" from a probe? Redwood Research/MATS 11.0 interns Taiga Someya (main contributor) and James Lucassen (supervisor) tested exactly this in a two-week sprint project.

They trained chameleons using the original authors' code and data, with abliterated Gemma-2-9B-it as the primary model, evaluated on GSM8K and MMLU. Key finding: when the probe monitors the concept the task requires, evasion succeeds but at enormous cost. Under the "mathematical" trigger, the math probe flags only 11% of correct GSM8K answers, yet accuracy falls from 89% (base model) to 46%. Chemistry and biology show the same trade-off. Capability drops 35–44 points under the relevant trigger versus only 18–21 points under the unrelated control trigger "elephant," indicating the cost is not generalized training perturbation but direct conflict with the task's required cognition.

The replication also revealed two discrepancies with the original paper. First, concepts the chameleons were never trained on (geography, physics, history) mostly remain visible to probes, contradicting the paper's claim that evasion generalizes to unseen concepts. Second, the paper's baseline MMLU and GSM8K scores match Google's published numbers for the pretrained Gemma-2-9B rather than the abliterated Gemma-2-9B-it that was actually fine-tuned, meaning the original paper understated the training cost.

The authors tested a larger model (31B), twice the training data, and a non-abliterated model; none eliminated the trade-off. The 31B model appears to hide math more cheaply (97% drops to 84%), but in logits against the base model its evading runs still lose about three-quarters as much math accuracy as the 9B (−1.9 vs. −2.4), and lose more on chemistry (−2.3 vs. −1.6).

The practical implication for AI safety monitoring: if a model must complete the very task being monitored (for instance, writing a backdoor while a probe watches for "backdoor"), current methods cannot evade the probe without substantial capability loss. The authors note that simple deception—where the model does not need to retain much cognitive capability—may be an exception, but more sophisticated violations are precisely the scenarios that demand capability. All experiment code is publicly available.

Voice AI company Deepgram processes quadrillions of audio tokens a year for more than 200,000 developers and 1,400 organizations. Its offline batch workloads—contact-center analytics, medical transcription and similar scenarios—are extremely sensitive to inference cost. In October 2026, Deepgram and the AWS Forward Deployed Engineering team published a complete engineering record of porting the Nova-3 speech-to-text model to Trainium2 chips, self-reporting that a single EC2 trn2.3xlarge instance processed 11–70% more audio per dollar.

The port was not a simple compilation swap. The team first ran Nova-3 in eager mode on Trainium2 via TorchNeuron and checked which operations fell back to the CPU, then compiled with torch.compile, producing separate static-shape graphs for the encoder and decoder. The compiled path ran every model operation on Trainium2 with no CPU fallbacks detected. Numerical accuracy was validated at two levels—layer-level output consistency against CPU and end-to-end Word Error Rate—ensuring alignment with Deepgram's reference values.

The real gains came from kernel-level optimization. The team adopted a flywheel strategy: enable an NKILib pre-built kernel, profile with Neuron Explorer, then tune or hand-write a replacement based on profiling insights. Three key findings each addressed a specific bottleneck. The reference kernel for the feed-forward network layer reloaded the full set of projection weights from off-chip memory for every token tile, making memory bandwidth the limiting factor as token count grew; the hand-tuned kernel loads each weight tile once and reuses it across multiple token tiles, achieving 2.0× lower latency at batch size 96. The experimental NKILib kernel for depthwise 1D convolution framed the operation as an implicit GEMM, leaving Tensor Engine utilization low; the team rewrote it as a broadcast multiply-accumulate loop on the Vector Engine, reaching full engine utilization for a 1.5× speedup at batch size 96. The conv2d layers hit a compilation bottleneck beyond batch size 32, preventing single-graph compilation; since NKILib had no conv2d kernel, the team wrote one from scratch in NKI, unlocking batch sizes above 32 and achieving up to 50× speedup on the convolution module at batch sizes ≤32.

The final source of gain was the runtime language. Nova-3 performs a sequence of small decoder calls for each audio chunk, making host-side dispatch the bottleneck at low batch sizes. Deepgram's production serving stack is itself written in Rust, so the team built a lightweight Rust runtime: Rust coordinates eight model replicas through a shared work-stealing queue, while a C layer loads and executes ahead-of-time-compiled NEFF files directly through the Neuron Runtime API, entirely bypassing the Python interpreter. Compared with a multiprocessing Python baseline, the Rust runtime eliminates interpreter overhead; compared with a multithreaded baseline, it further alleviates GIL serialization. At the throughput-maximizing batch size, the Rust runtime delivered a 7.5% throughput gain over the multiprocessing Python baseline.

It should be emphasized that all the above figures are self-reported measurements by Deepgram and AWS. The baseline is the compute options Deepgram had previously been running, whose specific configurations are not disclosed. The wide 11–70% range covers different workloads and batch sizes; readers evaluating their own scenarios should not directly extrapolate from this range.