raylad

Doesn’t work on mobile Safari latest version either. At least not when using iCloud private relay.

Completely failed if you search by Artist name

How do you keep the caches valid?

Really? San Francisco? Candy crush is produced by a Swedish company.

An obvious flaw in this study is that the meals were not anything like equivalent.

The minimally processed food consisted mainly of cheese, egg, and pieces of whole fruit: No unbound simple carbohydrates.

The ultra processed food included jelly, turkey, chips, a cookie, cereal, instant mashed potatoes: Lots of unbound simple carbohydrates and much less fiber.

Of course insulin would spike higher with such a meal. There’s nothing at all non-obvious about that.

Also, the protein wasn’t equivalent. The turkey roll is meat protein and cheese is dairy protein. Also, the salt content was probably higher in the ultra processed foods, and that was not controlled for.

If this could be replicated using exact equivalents that are either processed or not processed, I would be more convinced. But I really doubt that it would replicate.

There’s nothing surprising about this for people who have done casual mobile games (the genre including Candy crush).

What we do for those is to instrument them such that we know what levels are more difficult, what levels cause more users to abandon the game temporarily, or permanently, etc.

We then dynamically tune these levels until they at least no longer cause abandonment and ideally until they increase engagement.

My take away from this is that perhaps it’s better to not compact and instead to start new sessions whenever possible. I tell the agents to write a HANDOVER.md file that contains sufficient information to resume but not any extraneous information. That can be easily reviewed because it’s usually fairly short. Then, /clear and tell the new session to read that file and continue.

I just checked that model (ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S) and it does much much worse on the "Please recite Jabberwocky" test than the original bf16 does.

The bf16 only misses "snicker-snack" and this quantization becomes confused after the first stanza.

How does that compare with the bf16 version?

For my "Please recite Jabberwocky" test the bf16 almost passes but the ternary and even fp8 versions fail badly.

Wait till they start mixing every language in one sentence, and using poetic allusions that only scholars of each language would know.

Sounds great except it’s too tall to get under the bed.

In 5 years the models will probably be so much better and more compact that phones will be running models equivalent at least to Opus 4.6 if not Fable, at least within the areas they are tuned for (which probably won't include coding).