DEV Community: connerlambden

The GMAT redesign: three sections, one stubborn question

connerlambden — Mon, 24 Aug 2026 20:31:52 +0000

The GMAT does not look like it did in 2023. The familiar 200-800 scale is gone, replaced by 205-805. The Analytical Writing section now stands alone and unscored. Data Sufficiency, the section everyone dreaded, is folded into the same scored pool as Problem Solving. The exam is about an hour shorter.

And the two questions people actually care about are the same two they asked ten years ago. Is the score coachable? And what does the number really predict?

What changed

The current GMAT sits on a single scored section: Quantitative Reasoning, Verbal Reasoning, and Data Insights. Three sections, scored 60-90 each, combined into a total of 205-805.

Data Insights is the section with no direct ancestor. It combines data sufficiency, table and graph analysis, and reasoning from numbers. It is the most reasoning-heavy thing the GMAT has ever put on a scored section, because it rewards speed at reading a table and deciding what is sufficient, not speed at arithmetic.

What the score predicts

This is the part the ads do not want to linger on: the GMAT is a serious predictor. Study after study finds higher GMAT scores correlate with stronger performance in the first year of an MBA program, and MBA programs lean on the score in admissions because that relationship is real.

That is the thing to hold onto when someone tells you prep is a scam. The score measures something. It is not measuring your worth; it is measuring a specific blend of reasoning speed, composure under time, and judgment about what you can conclude from data. All three are learnable, and all three improve with deliberate practice.

What prep actually buys you

Here is where most marketing overshoots.

The dramatic score jumps in the ads are real but cherry-picked. The people in them are usually retakers who had a bad first sit, or people who switched to the new content late. Controlled gains from real practice are solid, modest, and they cluster in the same narrow window: 20 to 50 points for most people who work steadily. That is a real edge. It moves the needle for admissions odds at most schools, and it is not the transformation the billboards sell.

The faster, more durable win is exactly the one prep marketing has the least incentive to advertise: practice the reasoning, not the content. Data Insights rewards reading a table and deciding with sparse evidence. Verbal rewards telling an assumption from a fact. Both are trainable judgments, and both respond to volume with feedback far more than to another lecture.

The trap

The most expensive mistake a 2026 GMAT candidate can make is studying the old test. Huge amounts of online material are written for the pre-2024 exam, and most of it is now wrong about the format. If your study plan still has a dedicated "Data Sufficiency" block because a tutor you like said so, you are spending time on a configuration the test does not run.

The sharpest prep move is to ground everything in the current exam, use only material written for the new format, and point your practice at Data Insights and Verbal — the two sections easiest to improve and toughest to fake.

What actually helps

Keep it short, because it is short.

Practice timed, every session. The GMAT is a speed test wearing a reasoning disguise.
Review the miss harder than the hit. The score-feedback loop is the whole game.
Drill Data Insights daily. It is new, it is reasoning-heavy, and it is the easiest place to pick up points nobody is defending.

The through-line

The redesign did not make the GMAT more coachable. It made the test more honest: three sections, less content to hoard, more reasoning to train. That is good news for a reasoning practice that does not promise to make you smarter, only sharper at the specific judgment these sections actually measure.

If you want a free daily reasoning warm-up tuned to the exact judgment the GMAT and GRE verbal sections reward, the GMAT guide on IntelligenceMax digs into what the redesign means and where the coaching hype overshoots.

Disclosure: I founded IntelligenceMax, an adaptive reasoning practice. If any of this reads soft, say so.

Is the LSAT coachable? What the 2024 redesign settled

connerlambden — Mon, 24 Aug 2026 19:14:05 +0000

In August 2024 the LSAT dropped Logic Games. Studiers lost a whole section of the test overnight, their courses lost a quarter of their material, and the marketing copy went quiet for about a week before pretending it had always been this way.

The change is worth sitting with, because it quietly settled a claim people have argued about for years: is the LSAT a trainable skill, or a raw-aptitude measure you either have or don't?

What actually changed

The LSAT is now two Logical Reasoning sections, one Reading Comprehension section, and one unscored experimental section. Still scored 120 to 180. The writing sample is separate and unscored.

That one structural edit moves a lot of weight. Logical Reasoning is now two-thirds of the scored test. The section everyone treats as "LR is where the score lives" is now most of the exam by definition.

Coachable, but not in the way the ads mean

Here is the honest position, and it is a more interesting one than either side of the old argument gives you.

The LSAT is the most coachable of the big admissions tests. Not because a magic method exists, but because what it tests is unusually specific. Assumption, flaw, strengthen, weaken, inference: these are learnable moves, not reflexes you are born with. Get good at naming the job of a question stem on sight and you buy back real speed. That is real.

But "coachable" does not mean "15 points in a month." The dramatic scored jumps you see advertised are selected samples, shown because they are dramatic. Controlled gains from real practice are solid, but modest and clustered: they show up as a handful of scaled points and, more durably, in how you read under time. Call it a real edge, not a transformation.

The trap that costs you time

The most expensive mistake a 2026 LSAT studier can make is studying the old test. Huge amounts of reputable material are still geared toward Logic Games that no longer exists on it. If your schedule still has a Logic Games block, you are rehearsing a test that stopped being real in 2024.

The sharpest reprioritization you can do is to point almost all of your reasoning practice at the two Logical Reasoning sections, because that is where two-thirds of your score now comes from.

What actually helps

This part is not complicated, and it is worth saying plainly:

Know your question stems cold. Learn to recognize the job before reading the stimulus, so you read the argument with a purpose instead of hoping.
Read comparatively. The comparative passage set rewards tracking who argues what. Practice on real dense prose, not warmed-over summaries.
Drill timed, review the miss. The feedback loop is the highest-yield activity there is. Get one wrong and understand exactly why.
Retake with a plan. A focused repeat beats passive repetition.

The through-line

The design of the new LSAT matches what the evidence has said all along: you get better at the reasoning you actually practice, in the form you practice it. You do not get smarter in general; you get sharper at this specific, high-value judgment.

That is the whole honest frame. And it is why, if you want a daily reasoning warm-up that chases the exact distinctions the LSAT cares about, there is one with no score promise: IntelligenceMax's LSAT guide, which breaks down the current test format, what prep buys you, and where the hype overshoots.

Disclosure: I founded IntelligenceMax, an adaptive reasoning practice. If any of this reads soft, say so.

Does learning formal logic make you better at reasoning? What the evidence actually says

connerlambden — Mon, 24 Aug 2026 17:31:56 +0000

Will studying logic make you better at reasoning? is one of those questions that sounds naive until you look at how long people have argued about it. In 1917, Edward Thorndike tried to settle it with his doctrine of identical elements: training transfers only to tasks that share parts with what you practiced. A century of replication attempts later, the answer has not gotten more flattering for the optimists.

Here is the honest state of the evidence, written for a developer or self-learner who wants to reason better and distrusts both brain-training hype and nothing-transfers cynicism.

What does not hold: far transfer to general intelligence

Meta-analyses of cognitive training find consistent near-transfer gains. Broad far-transfer to general intelligence is weak or null once you control for placebo and reporting bias:

Simons et al. (2016), Do Brain-Training Programs Work?, found limited evidence of far transfer beyond the trained task.
Melby-Lervaag, Redick and Hulme (2016) found near-zero transfer to general ability for working memory training.
Sala and Gobet (2017) found practice effects confined to the trained task across many training types.

Translation: studying formal logic can make you better at formal logic, written argument evaluation, and certain math. There is no credible lane from a month of truth tables to getting broadly smarter.

What does hold: near transfer and honest practice

The transfer literature is not a wall. It is a corridor:

Skills transfer along shared structures, not shared surfaces. Train the exact inference you want and you get better at it. This is why LSAT drilling works for LSAT logical reasoning, and algebra drills help a math exam.
Practice with feedback beats passive exposure. You learn to evaluate arguments by evaluating hundreds of them and seeing why you were wrong, not by reading textbook definitions of validity.
Knowledge compounds. The best reasoners know a great deal about the thing they are reasoning about, plus they have practiced a small set of moves: premises, support, warrants, and a sharp eye for the two choices in a dilemma.

The practical upshot for a builder

This evidence is why IntelligenceMax exists. It is a reasoning practice that does not claim to raise your IQ, and says so in the product. It drills the narrowest useful judgment of all: what is actually supported, with what strength, by this premise? delivered adaptively, scored honestly, with feedback on every answer.

Questions are generated live so the gym never goes stale, and you always see fresh items at your current boundary.
The score is an item-response-style estimate of current ability on that task, with an honest uncertainty band, not an IQ certificate.
The design assumption is near transfer: train the judgment you want, in the form you want it in.

The short answer to the title question

Yes, but only in ways that survive scrutiny:

It improves reasoning on tasks that share structure with practice.
It does not reliably raise a score on a reasoning measure far from the practice.

Anyone selling you claim two without claim one is either describing older, better-trodden practice (law, math, engineering, code) or overselling your future.

See how we score reasoning and where we draw the claim boundaries on the science page at intelligencemax.ai. And if this reads soft anywhere, say so. I am trying to train exactly the judgment this piece exercises.

If you want to see how one product translates this into an adaptive practice with honest scoring and no IQ claims: intelligencemax.ai/science

Honest edtech claims: what reasoning practice can and cannot do

connerlambden — Mon, 24 Aug 2026 15:22:21 +0000

Edtech fails in two directions. It overclaims, selling a few games as a way to boost intelligence and reshape a life. Or it underclaims, staying so vague that it says nothing at all. Both are dishonest, and both erode trust in the whole category.

The transfer evidence is clear

The training-transfer literature is settled enough to build on. Melby-Lervag and Hulme (2013) meta-analyzed working-memory training and found reliable gains on trained tasks but little transfer to general cognition. Sala and Gobet (2017) reviewed reasoning training and reached the same conclusion. A 2022 Translational Psychiatry trial gave middle-aged adults eight weeks of adaptive n-back training and found strong practice effects with no near or far transfer, at the cognitive level or in brain structure.

The pattern across all of them: you improve at what you train, and sometimes at things closely related. You do not measure broadly. That is not a product failure. It is what the evidence actually shows.

What honest edtech looks like

An honest reasoning product claims a narrow edge. It trains a specific skill, adapts to your level, and scores you the way a psychometric instrument does. It does not claim to raise your IQ, fix your career, or make you a genius.

At IntelligenceMax I built practice scored with item response theory. The model estimates a latent ability behind your answers and calibrates difficulty to it. That is not magic. It is the same math behind standardized tests. What it buys is honesty: we can score practice fairly and measure the right thing, without promising to transform your mind.

Honest claims are the strategy

People do not trust brain training. The instant a product promises a twenty-point IQ jump, everyone discounts everything else it says. Trust survives when the claims do. We prefer narrow ones.

If reasoning practice has a future, it is the narrow kind: getting better at telling what is supported from what is merely plausible, reading closely, spotting hidden assumptions. That is useful on its own. It is also the only claim the research supports.

Why we score reasoning practice with Item Response Theory (and what it does not claim)

connerlambden — Sat, 22 Aug 2026 16:04:33 +0000

When I started building IntelligenceMax, I faced a question that every cognitive training product has to answer: how do you measure whether someone is getting better at reasoning?

The obvious answer is "track their accuracy." The obvious answer is wrong.

The problem with raw accuracy

If a learner gets 8 out of 10 questions right on day 1 and 9 out of 10 on day 10, did they improve? Maybe. But you cannot know without knowing the difficulty of those questions. Easy questions inflate accuracy. Hard questions depress it. Raw accuracy confounds ability with the difficulty of the specific questions you happened to ask.

This is why standardized tests do not report raw accuracy. They report scaled scores estimated from the difficulty of the items answered. The SAT does not tell you "you got 47/58." It tells you "your score is 720," where 720 is an estimate of your ability that accounts for which questions you got right and wrong.

What IRT actually does

Item Response Theory (IRT) models the probability that a person with a given ability level answers a given item correctly. The simplest model, the 1PL or Rasch model, uses one parameter per item: difficulty. The probability of a correct answer is a logistic function of the difference between the person's ability and the item's difficulty.

$$P(\text{correct}) = \frac{1}{1 + e^{-(\theta - b)}}$$

Where θ is the person's ability and b is the item's difficulty. Higher ability relative to difficulty means higher probability of a correct response.

The 2PL model adds a discrimination parameter. The 3PL model adds a guessing parameter. For our use case (LLM-generated MCQs where guessing is not the primary concern), we use a 2PL model.

The benefit: ability estimates are comparable across different sets of questions. If you answer 7 hard questions correctly and I answer 9 easy questions correctly, IRT can tell you your ability is higher than mine, even though my raw accuracy is higher.

Adaptive difficulty

IRT also enables adaptive testing. Once you have an ability estimate, you can select the next question to maximize information about that person's ability. For a 2PL model, the most informative question for a person at ability θ is one where the difficulty b is close to θ.

This is what CAT (Computerized Adaptive Testing) does. The GRE and GMAT use it. We use a simplified version: generate questions at a target difficulty based on the current ability estimate, and adjust after each response.

The benefit for the learner: every question is at the edge of their ability. Not too easy (boring), not too hard (frustrating). This is the zone of proximal development, operationalized psychometrically.

What IRT does not do

Here is where I want to be careful, because the cognitive training industry has a history of overclaiming.

IRT does not measure intelligence. It measures performance on a specific set of items. If those items are reasoning MCQs, IRT estimates your ability at answering reasoning MCQs. It does not estimate your general intelligence (g). Ability on a specific reasoning task and general intelligence are correlated but not identical.

IRT does not prove transfer. Getting better at our reasoning MCQs, even with adaptive difficulty and accurate scoring, does not mean you have improved at other reasoning tasks. The evidence for far-transfer from cognitive training to general cognition is weak (Melby-Lervåg & Hulme, 2013; Sala & Gobet, 2017). We do not claim our platform improves general cognitive ability. We claim it provides adaptive practice at a specific skill, scored with a method that accounts for item difficulty.

IRT does not make claims about IQ. IQ is a specific construct measured by specific standardized tests. Our ability estimate is not an IQ score. It is a performance estimate on our specific item pool. Conflating the two is what brain training companies do, and it is why the field has a credibility problem.

Why we chose IRT anyway

Because it is the honest way to measure what we are doing. We are providing practice at a specific reasoning skill (distinguishing supported arguments from plausible ones). To know whether that practice is working, we need to measure ability changes over time. Raw accuracy cannot do that because question difficulty varies. IRT can, because it separates ability from difficulty.

The alternative is to not measure at all, or to report raw accuracy and pretend it means something. Neither is honest.

The engineering trade-off

IRT requires item parameter estimation. You need enough responses per item to estimate difficulty and discrimination. For a platform generating fresh questions with LLMs, this is a challenge: each question is new, so there are no historical parameters.

Our approach: we generate questions at a target difficulty (determined by prompt engineering), then calibrate item parameters from response data as it accumulates. Early on, we rely on prompt-specified difficulty. Over time, the data corrects.

This is a known trade-off in adaptive testing with generated items. The alternative is to use a static item bank with pre-calibrated parameters, but that sacrifices freshness and variety. We chose variety over parameter precision, and we are transparent about that.

The bottom line

IRT-based scoring is a tool for honest measurement of a specific skill. It does not make IntelligenceMax a brain training app, an IQ test, or a general intelligence improver. It makes it a platform that practices one reasoning skill, measures performance on that skill accounting for item difficulty, and adapts difficulty to the learner.

That is a narrow claim. I think narrow claims are the honest ones.

The LSAT-to-1L reasoning carryover is real, but narrower than people think

connerlambden — Sat, 22 Aug 2026 02:51:41 +0000

I built an adaptive reasoning practice app, so I spend a lot of time thinking about what transfers from one reasoning-heavy task to another. The LSAT-to-1L question is the one I get most often, and it's the one where the popular answer is most wrong.

The popular claim: LSAT prep makes you better at law school because law is just logic.

The honest claim: LSAT Logical Reasoning trains a specific, narrow skill (spotting the distinction between an answer that's supported and one that's merely plausible), and that skill does carry into 1L exam writing, but only as one component among several. It does not replace learning doctrine, and it does not make law school easy.

What actually transfers

The LSAT LR section is built around a core move: read a short argument, identify its gap, and then pick the answer that either exploits or patches that gap. The wrong answers are designed to be tempting, which means they're plausible-sounding but unsupported. The right answer is the one that does the work the question asks.

That distinction, supported vs. plausible, is the same move that separates good 1L exam answers from average ones. On a law school exam, the professor gives you a fact pattern and asks you to identify the claims, defenses, and counterarguments. The students who do well are the ones who can look at a set of facts and figure out which rule is actually triggered, not which rule looks like it might be. They can distinguish a case where negligence applies from one where it looks like it applies but doesn't.

That's the transfer. It's real. It's the reason LSAT score correlates with 1L GPA more than undergraduate GPA does. But it's one component.

What doesn't transfer

The LSAT doesn't teach you doctrine. It doesn't teach you the elements of negligence, the rule against perpetuities, or the Erie doctrine. A 175 LSAT scorer who hasn't learned the material will still fail the exam. The reasoning skill is the scaffold, but the doctrine is the building.

The LSAT also doesn't teach you the exam-writing format law professors reward. IRAC isn't hard, but it's specific, and it's not intuitive if you've never seen it. A brilliant reasoner who writes a formless exam answer will score lower than a mediocre reasoner who structures it well. The LSAT trains reasoning, not legal writing.

And the LSAT doesn't teach you to read cases the way law school requires. Reading a case for the holding is a different skill from reading an LR stimulus for the flaw. Both are close reading, but the objects are different.

The narrow version of the claim

The honest version of the LSAT-to-1L carryover is this: if you trained the distinction-finding muscle for the LSAT, you'll find it easier to do the distinction-finding part of 1L exams. That's it. It's a head start on one skill, not a substitute for the rest of law school.

If you're a 0L and you're asking whether LSAT prep was wasted if you don't go to law school: no. The same distinction (supported vs. plausible) shows up in the GRE critical reasoning questions, in the GMAT Critical Reasoning section, and in the MCAT CARS section. It's a general reasoning skill, not a law-specific one. That's why standardized tests across fields all include some version of it.

The practical takeaway

If you're heading to law school and you already did LSAT prep, the thing to do is not to rest on it. The reasoning head start is real, but it decays if you don't use it, and it's only one part of the exam. Spend your 1L energy on doctrine and exam structure. The reasoning will be there when you need it.

If you're a 0L who hasn't taken the LSAT yet, the highest-value thing you can drill is the LR section, because the distinction-finding skill compounds across the whole test and then carries forward. Reading comp matters, but it rewards volume. LR rewards learning the structural move, and that move transfers.

I'm Conner. I build IntelligenceMax, an adaptive reasoning practice app that generates LSAT-style and GRE-style questions from frontier LLMs and scores them with item response theory. It's not a test-prep replacement and it won't raise your IQ; it's a reasoning gym for the distinction-finding muscle I'm talking about above.

How to improve your IQ (without lying to yourself)

connerlambden — Mon, 20 Jul 2026 16:54:42 +0000

Search demand for "how to improve your IQ" and "how to improve your intelligence" usually wants one of three different things, and sellers often blur them.

Three claims people mix up

Test score movement. Practice effects, item familiarity, sleep, and anxiety can move an IQ score without proving a durable jump in general intelligence (g).
Competence growth. Crystallized skill and domain expertise keep rising for years with hard study. That is real improvement even when Raven-like fluid scores barely budge.
Far transfer from brain games to broad Gf. Near transfer to trained tasks is common. Large lasting Gf gains are not the default finding once active controls enter the picture.

If an article promises a free upgrade to general intelligence from puzzles alone, shrink the claim until it matches the evidence.

A boring adult plan that survives contact with research

Sleep and aerobic exercise first.
Deliberate practice in a hard domain (math, language, code, music).
Optional adaptive reasoning practice if you enjoy it.
Treat any practice estimate as training feedback, not a clinical IQ diagnosis.

Sources I maintain

Claim-size map: intelligencemax.ai/guide
Evidence pack: osf.io/kja9b

Disclosure: I build IntelligenceMax, an adaptive reasoning gym. Practice software is not a promise that apps raise trait IQ.

Lumosity, Elevate, Peak: same aisle, different drills

connerlambden — Thu, 16 Jul 2026 21:06:32 +0000

Search demand lumps Lumosity, Elevate, Peak, BrainHQ, and NeuroNation into one bucket: "brain training apps." Product design does not.

Three different product jobs

Language / productivity drills (Elevate-shaped): reading, writing, listening, math fluency. Progress charts track those drills.
Cognitive mini-games (Lumosity / Peak-shaped): memory, attention, flexibility, speed under time pressure.
Reasoning practice (what I ship): adaptive multiple-choice distinctions at your edge, with an honest practice score that is not a clinical IQ claim.

If you evaluate all three with the same KPI ("raises IQ"), you will misread the literature and mis-buy the subscription.

What the evidence usually supports

Independent reviews of commercial brain training typically find:

Near transfer: you get better at the trained tasks and close cousins.
Far transfer: weak under careful controls for everyday reasoning / general intelligence.

Useful anchors: Simons et al. 2016; Melby-Lervåg, Redick & Hulme 2016 on working-memory training. Pick the app whose drills match the skill you actually want to practice.

I keep a public claim-size map here: intelligencemax.ai/guide (OSF: osf.io/kja9b).

Why this matters for builders

If you are shipping adaptive difficulty, do not market "adaptive" as "clinically validated IQ gains." Adaptive means the item bank tracks an ability estimate for your task family. That is useful. It is still not an IQ test.

Disclosure: I build IntelligenceMax, an iOS reasoning gym. Practice estimate ≠ clinical IQ.

Quick chooser

Want…	Prefer…
Grammar / reading fluency games	Elevate-like
Short attention/memory mini-games	Lumosity/Peak-like
Hard MCQ reasoning under adaptive difficulty	Reasoning gym

Same aisle. Different drills. Claim size is the product.

What "adaptive" should mean in a reasoning practice app

connerlambden — Thu, 16 Jul 2026 20:58:45 +0000

Most apps that say adaptive just mean a streak counter and a harder level after three greens.

I shipped IntelligenceMax on the App Store with a narrower definition.

Adaptive as serving, not as a vibe

After each answer, the app updates a practice estimate and uses that to choose later item difficulty. Explanations appear after submit, not before. That is the loop: miss → see why → get something near that edge again.

What the score is not

The estimate uses an IRT-shaped update and can be shown on an IQ-looking scale for readability. It is still an in-app practice metric. It is not a normed clinical IQ battery, and shipping to the App Store does not change that.

Why builders should care about the claim size

App Store categories reward big cognitive verbs. Independent reviews of commercial brain training are usually clearer about near transfer than about far transfer. If your product only proves "people get better at your items," say that. Do not let the marketing department invent adult g.

Try / read

App Store: https://apps.apple.com/us/app/intelligencemax/id6786447264
Methods: https://intelligencemax.ai/science

I build IntelligenceMax. Feedback on explanation quality beats applause for IQ theater.

I graded my own ML option forecasts. Here's the Brier score.

connerlambden — Tue, 14 Jul 2026 22:24:55 +0000

In May I promised — out loud, on the internet, where people can screenshot you — that I'd grade Helium's published prob_itm forecasts when the June 2026 AAPL contracts expired.

June 26 came. The contracts died. So I graded them.

This is not a victory lap. Mean Brier loss on n=2 is 0.3846. A coin flip that always says 0.5 scores 0.25 when the world resolves to 0 or 1. We did worse than a coin flip. That is information.

What we froze in May

Contract	Helium `prob_itm`	Market-ish implied
AAPL $310C 2026-06-26	0.42	~0.50
AAPL $295P 2026-06-26	0.23	~0.24

What the stock actually did

AAPL NASDAQ close on 2026-06-26: $283.78

$310 call → OTM → realized ITM = 0 → Brier = (0.42 − 0)² = 0.1764
$295 put → ITM → realized ITM = 1 → Brier = (0.23 − 1)² = 0.5929

Honest read (short)

The call forecast was closer to reality than the market's ~0.50 shrug — but "closer" is not "good." Any non-zero probability on an OTM finish still costs you.
The put is the bruise. Probabilities looked aligned pre-expiry; then the underlying fell hard enough that both market and model had understated the ITM risk. Helium's number was far from 1, so the Brier loss is ugly.
n=2. This is a promise kept, not a calibration paper. If you want to treat it as science, wait for more expiries on the honesty board.

Reproduce

git clone https://github.com/connerlambden/helium-mcp-cookbook
cd helium-mcp-cookbook
python calibration/grade_june_expiry.py

Frozen inputs live in calibration/june_2026_forecasts.json.

Why bother publishing a bad score

Because the alternative is the usual ML options blog post: a backtest, a vibe, a screenshot of a green P&L. Publishing prob_itm with a fixed expiry is a tiny discipline. Grading it after the fact is the second half of that discipline.

If you log your own forecasts with recipe 02, you can run the same scorer when your contracts expire. I'll keep adding rows.

—

Also adjacent, if you're here for news rather than options: we published a free 212×37 news-outlet framing corpus with an explorer. Different animal. Same "put numbers where people can grade them" instinct.

An IRT-shaped practice score is not an IQ test

connerlambden — Tue, 14 Jul 2026 22:10:27 +0000

IntelligenceMax updates a learner estimate after every answer and displays it on the familiar IQ scale. The arithmetic resembles item response theory. That resemblance is useful, but it is easy to ask the formula to carry more meaning than it can bear.

A formula can be internally consistent while its inputs remain uncertain. Here, the distinction begins with the item parameters.

The update rule

For ability θ, item difficulty b, and discrimination a, the model assumes a logistic P(correct). The first estimate has a normal prior centred at 100 with SD 15. After an answer, the implementation takes one capped local update using score and Fisher information. Per-answer change is capped at 1.5 points.

The calculation is the easy part. The inputs are where the uncertainty lives.

Model-assigned is not empirically calibrated

The AI that generates each question also assigns difficulty and discrimination. Those values are checked for shape and range, but they are not fitted from a norming sample. Calling the update IRT-style describes the logistic form. It does not establish that 115 means the same thing as a score of 115 on a validated instrument.

What this number can support

Within those limits, the estimate can support an adaptive practice loop. It cannot establish a clinical Full Scale IQ, a permanent change in general intelligence, transfer to school or work, or treatment of a cognitive condition.

Reviews of commercial brain training find the strongest evidence on trained tasks and close relatives. Broad gains are less convincing against active controls (Simons et al., 2016; Melby-Lervåg et al., 2016).

Until empirical calibration exists:

This is a logistic, IRT-shaped practice estimate that uses model-assigned item parameters. It is not a normed IQ score.

Public notes: https://intelligencemax.ai/science
Evidence map: https://intelligencemax.ai/guide
OSF: https://osf.io/kja9b/

Disclosure: I built IntelligenceMax. The technical question I most want readers to challenge is whether useful empirical calibration is possible for generated, mostly one-off items without quietly turning the system into a fixed bank.

How to improve your IQ without lying to yourself

connerlambden — Tue, 14 Jul 2026 21:02:57 +0000

People type “how to improve IQ” into Google the way thirsty people type “how to hydrate instantly.” Fair. Also: the internet will sell you a miracle before it sells you sleep.

What IQ is (and is not)

IQ is a test score. Useful. Incomplete. Noisy day to day. It is not a moral ranking, a destiny, or a personality.

If your plan is “raise my number by Friday,” you are shopping for fog. Scores move with sleep, anxiety, practice effects, and which test you took.

What actually tends to help

Schooling and hard domain practice still look better than most commercial brain-training promises. Exercise and sleep help cognition more reliably than another puzzle leaderboard. Reading hard books and taking feedback that can embarrass you remains undefeated.

Near transfer vs far transfer

Near transfer is real: practice a skill, get better at that skill and nearby ones.

Far transfer — train a game, raise general intelligence forever — is where ads get loud and meta-analyses get quiet. Simons et al. (2016) on commercial brain training. Melby-Lervåg and colleagues on working-memory training. Treat “this app raised my IQ 15 points” with the same skepticism you would bring to a late-night fitness infomercial.

A plan that is kinder than chasing points

Pick one hard skill you will actually use.
Force weekly output that can be graded or rejected.
Sleep like it is part of the training.
Revisit IQ only if you need it for something concrete. Do not let it narrate your identity.

Disclosure

I build IntelligenceMax: adaptive reasoning practice. We do not claim it raises IQ. We care about transfer honesty.

Longer map: intelligencemax.ai/guide

OSF notes: osf.io/kja9b

Canonical essay: How to improve your IQ (without lying to yourself)

Getting sharper is allowed. Lying to yourself about what transferred is optional, and expensive.