The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →The argument that memory, not raw scale, is the next step for large language models is the author’s thesis, and the 3M-token experiment behind it is a single-system test on a synthetic corpus. The result is a strong demonstration that one compression-and-recall design can retrieve planted facts from a very large, distracting history. It does not show that memory is the consensus next frontier, or that the approach generalizes to other models or to ordinary conversations.
What the author is arguing
The article, published on DEV Community under the byline uos1231234, contends that gains in model capability are not explained by scale alone. Its proposal is that a system should compress prior interaction into useful state, then retrieve specific details when that state turns out to be incomplete. The author states the view directly:
As an Amazon Associate I earn from qualifying purchases.
“In my view, the LLM’s next step should be memory — giving LLMs a human-like memory mechanism instead of only an attention mechanism.”
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →That sentence is an opinion, not a validated technical conclusion. The model used in the work is reported as deepseek-v4-flash, accessed through a tao-deepseek relay, and the compression, archival and recall steps were run by a system agent using the same model. The author’s setup is therefore one model configuration in one implementation.
#1 Best Overall
- MADE FOR THE MAKERS: Create; Explore; Store; The T7 Portable SSD delivers fast speeds and durable features to back up any endeavor; Build your video editing empire, file your photographs or back up your blogs all in an instant
- SHARE IDEAS IN A FLASH: Don’t waste a second waiting and spend more time doing; The T7 is embedded with PCIe NVMe technology that brings fast read and write speeds up to 1,050/1,000 MB/s¹, making it almost twice as fast as the T5
- ALWAYS MAKE THE SAVE: Compact design with massive capacity; With capacities up to 4TB, save exactly what you need to your drive – from large working files to game data and everything in between
- ADAPTS TO EVERY NEED: Whether using a PC or mobile phone, count on the T7 for extensive compatibility²; It’s a true team player when it comes to heavy-duty application usage or file-saving
- HI RESOLUTION VIDEO RECORDING: Record Ultra High Resolution (4K 60fs) videos directly onto the T7 Portable SSD with your favorite camera or mobile devices; Supports iPhone 15 Pro Res 4K at 60fps video and more³
What the experiment actually measured
The report, titled “MRCR-3M Constrained Recall Experiment — Data Analysis Report (2026-09-23),” evaluates the whole context-engineering system rather than one part of it. That system combines tiered compression, envelopes, tombstones and a recall fallback. The report does not isolate how much each component contributed, so the headline number describes the package.
The corpus is large in token-equivalent terms, which is different from a 3-million-token context window being filled with live conversation. The figures below are the report’s own.
| Element | Value reported | Qualification |
|---|---|---|
| Total size | About 3,007,411 token-equivalents | Author’s count; a local estimate, not a provider tokenizer figure |
| Blocks | 64 blocks, about 118.5K characters each | Synthetic corpus built for this test |
| Distractors | 384 same-distribution distractor blocks | Designed to look like the target material |
| Targets | 128 golden service-to-key mappings | Planted facts the system must recover |
| Primary scoring | A key string appears in the reply | Exact-string match |
| Stricter scoring | Service and key must appear on the same line | Described by the author as a tighter pairing check |
Because the corpus is synthetic and the scoring is exact-key, a high score says the system can find and state planted strings. It says less about whether the system understands them.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #2
- Get NVMe solid state performance with up to 1050MB/s read and 1000MB/s write speeds in a portable, high-capacity drive(1) (Based on internal testing; performance may be lower depending on host device & other factors. 1MB=1,000,000 bytes.)
- Up to 3-meter drop protection and IP65 water and dust resistance mean this tough drive can take a beating(3) (Previously rated for 2-meter drop protection and IP55 rating. Now qualified for the higher, stated specs.)
- Use the handy carabiner loop to secure it to your belt loop or backpack for extra peace of mind.
- Help keep private content private with the included password protection featuring 256‐bit AES hardware encryption.(3)
- Easily manage files and automatically free up space with the SanDisk Memory Zone app.(5). Non-Operating Temperature -20°C to 85°C
Results across the runs
The author reports a sequence of runs, with the scoring changing as the pipeline was repaired. Each row is out of 128 golden mappings.
| Run | Score | Share | What changed, per the author |
|---|---|---|---|
| r1 | 116/128 | 90.6% | Initial run |
| r2 | 118/128 | 92.2% | Early losses traced to envelope sections swallowed by a chunker |
| r3 | 114/128 | 89.1% | Not stated |
| r7 | 126/128 | 98.4% | Fence, retry, tombstone and reminder changes |
| r7-clean | 126/128 | 98.4% | Clean rerun of r7 |
| r8 | 128/128 | 100% | Final run; the last two recovered needles are attributed to key preservation during compression |
The author says the final question round used zero tool calls. That detail matters: it means the r8 answers did not depend on a recall tool call at answer time, which is the author’s reading of why the last two items were recovered. The author reports this as an interpretation of the runs, not as a controlled ablation.
The same report also lists 3,059 passing tests in its baseline. A later count gives 3,062 total tests, with one skipped, one todo and one stale environment failure. Those are software-health figures for the project, not evidence about memory quality.
Rank #3
- MADE FOR THE MAKERS: Create; Explore; Store; The T7 Portable SSD delivers fast speeds and durable features to back up any endeavor; Build your video editing empire, file your photographs or back up your blogs all in an instant
- SHARE IDEAS IN A FLASH: Don’t waste a second waiting and spend more time doing; The T7 is embedded with PCIe NVMe technology that brings fast read and write speeds up to 1,050/1,000 MB/s¹, making it almost twice as fast as the T5
- ALWAYS MAKE THE SAVE: Compact design with massive capacity; With capacities up to 4TB, save exactly what you need to your drive – from large working files to game data and everything in between
- ADAPTS TO EVERY NEED: Whether using a PC or mobile phone, count on the T7 for extensive compatibility²; It’s a true team player when it comes to heavy-duty application usage or file-saving
- HI RESOLUTION VIDEO RECORDING: Record Ultra High Resolution (4K 60fs) videos directly onto the T7 Portable SSD with your favorite camera or mobile devices; Supports iPhone 15 Pro Res 4K at 60fps video and more³
How the author explains the gains
The mechanism the author credits has three parts. First, compression keeps some information in a state summary. Second, envelopes and tombstones record lineage, so the system can tell what was compressed and when. Third, a recall fallback runs when the summary is incomplete. The author attributes most of the late gains to the first mechanism: information preserved by the compression path, not fetched by a later query.
That attribution is plausible but unverified outside this run. The report does not include a version of the system with the fallback removed or with compression turned off, so it cannot separate the contributions cleanly.
Where the result is weaker than it looks
The author documents several problems, and they are worth reading as part of the result.
Rank #4
- Easily store and access 2TB to content on the go with the Seagate Portable Drive, a USB external hard drive
- Designed to work with Windows or Mac computers, this external hard drive makes backup a snap just drag and drop
- To get set up, connect the portable hard drive to a computer for automatic recognition no software required
- This USB drive provides plug and play simplicity with the included 18 inch USB 3.0 cable
- The available storage capacity may vary.
- Token-count calibration. A local counter triggered a fixed 512K reminder early. The reminder said session context exceeded 512K tokens, while provider prompt-token readings in blocks 10, 12 and 13 were 202,650, 212,223 and 232,036, roughly 210K. The author attributes the gap to the local counter overestimating tokens on repetitive material.
- A race on mutable history. An index built from the length of the history became invalid when compression changed that history during the ASK interval. A fallback that scanned the history for the last assistant string reply rescued the result, so the failure was caught but not designed out.
- Answer contamination. A revival test initially replayed an existing answer byte for byte. The author removed the questions and answers to make a clean rerun, which means a rerun that reproduces an earlier answer is not evidence of recall.
- Counts and limits. The report records 61 retained tombstones from 51 compressions and 11 M3 batches. A 61-message mailbox was read in two pages, 50 and then 11. The author says a targeted probe had zero limit collisions and zero violations.
Recall is not the same as long-context reasoning
A 2026 paper in the ACL Anthology makes the broader methodological point. Retrieval-centric benchmarks can show that a model located a buried item without showing that it reasoned over long context, and they can be vulnerable to leakage, short-circuiting and setups that are easy to identify. The paper points to a wider set of tests: multi-hop inference, aggregation, and reasoning about information that is absent.
The author’s test is almost entirely of the first kind. Finding 128 exact keys across a distractor-heavy corpus is a useful retrieval check. Whether a memory system can combine facts across blocks, count them, or conclude that something is missing is a different question, and the report does not test it.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteHow to judge a memory claim
Memory systems can be compared on the same axes the report uses, which is more useful than comparing headline scores.
Best Value
- Capacity Display Variance: 500GB external ssd often appears as around 465GB on Windows. MacOS can show full 500 GB capacity. This is binary calculation difference and doesn’t affect SSD hard drive actual physical storage
- 1050 MB/s Speed: Instantly access to your files with blazing-fast 10Gbps external SSD read up to 1050MB/s and write up to 1000MB/s. LED Light indicates USB SSD instant activity
- Data Security: Solid state drives S.M.A.R.T. health diagnostics and adaptive TRIM optimizing data block management ensures consistent write speeds and extends the longevity of the portable SSD
- USB-C & USB-A Cable: Both cables featuring rapid USB 3.2 Gen2, this USB SSD effortlessly bridges devices, enabling seamless cross-platform file transfers and backup between computers, smartphones, tablets and iPhone
- Always Fast: No slowdowns for large file transfers. With SLC caching (25% of current available capacity allocated as high-speed cache), this external SSD delivers steady 10Gbps for transfers within the cache capacity
- What carries information: a compressed state summary, a raw archive that is searched later, or both.
- When recall runs: always, only when the summary is incomplete, or after a verification step.
- Evaluation breadth: exact-string retrieval alone, or also multi-hop reasoning, aggregation and absence detection.
- Reliability controls: lineage records, protection against contaminated answers, calibration of token counts, and handling of history that changes mid-run.
- Generalization evidence: one model and one system, or independent runs across models, datasets and task types.
On the last axis, this experiment is at the weak end. Its evidence is one system, one model configuration and one synthetic corpus. The author has not published independent replication, cross-model results, or performance on ordinary user conversations, and the report does not compare the design with alternative memory architectures.
What is established and what is not
The experiment establishes that one compression-and-recall pipeline, run with one model, recovered all 128 planted service-to-key mappings in its final run, and that the author documented the failures and fixes along the way. It does not establish that memory is the next step for LLMs in general, that scaling has stopped mattering, or that a 3M-token-equivalent corpus is a standard benchmark. For a reader deciding whether to take the thesis seriously, the useful next evidence would be the same test repeated on other models, a version with components switched off, and tasks that require reasoning over the recovered facts.
The 3M figure is a token-equivalent estimate from the author’s counter, not a measured context length from a model provider, and that distinction is the one readers most often lose when they see the headline.
Recommended Free Tools
The Bottom Line
The memory thesis is a reasonable hypothesis supported by one carefully documented but narrow experiment. The 128/128 result shows that a compression-plus-recall system can retrieve planted keys in a synthetic corpus; it does not yet show that memory is the field’s next step or that the design works beyond this setup.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
