41554 blogs · [ { "id": "01a087e3-4d12-73f8-8382-32138f77f585", "title": "unyolo.io fixes this", "url": "https://solmaz.io/x/2087789802049404981/", "published_at": "2026-08-13T06:34:12+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32138f837450", "title": "😎", "url": "https://solmaz.io/x/2087776814651338815/", "published_at": "2026-08-13T05:42:36+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32138ff1784c", "title": "A similar thing can be done with codex session compaction summaries as well btw", "url": "https://solmaz.io/x/2087484201419334104/", "published_at": "2026-08-12T10:19:51+00:00" }, { "id": "01a087e3-4d12-73f8-8382-321390c689c7", "title": "This type of estimations are useful not only for local but for all inference providers", "url": "https://solmaz.io/x/2087382719940501673/", "published_at": "2026-08-12T03:36:36+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139174554c", "title": "mfw trying not to think about the 100 models that dropped this week", "url": "https://solmaz.io/x/2086983068959879561/", "published_at": "2026-08-11T01:08:32+00:00" }, { "id": "01a087e3-4d12-73f8-8382-321391a53fe0", "title": "Deterministic monitors make long-running agents reliable", "url": "https://solmaz.io/x/2086844493672980650/", "published_at": "2026-08-10T15:57:53+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139294a24d", "title": "Apparently this went to HN front page briefly last night, so sharing it here as well", "url": "https://solmaz.io/x/2086753216981770308/", "published_at": "2026-08-10T09:55:11+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139297f810", "title": "Local AI will favor LPDDR and MoE", "url": "https://solmaz.io/x/2086715364642365804/", "published_at": "2026-08-10T07:24:46+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139372ad2f", "title": "urgh typo, I meant 35-40 tok/s", "url": "https://solmaz.io/x/2086404453222298076/", "published_at": "2026-08-09T10:49:19+00:00" }, { "id": "01a087e3-4d12-73f8-8382-321393becaf9", "title": "Apparently Alibaba did all this work, published a paper, but did not create a public backup of...", "url": "https://solmaz.io/x/2086392908463382897/", "published_at": "2026-08-09T10:03:27+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139495991a", "title": "I have made an update to my theoretical upper bound calculation to also predict prefill speed", "url": "https://solmaz.io/x/2086346067952677215/", "published_at": "2026-08-09T06:57:19+00:00" }, { "id": "01a087e3-4d12-73f8-8382-321394b61201", "title": "Theoretical limits can expose inference problems", "url": "https://solmaz.io/x/2086333655119720782/", "published_at": "2026-08-09T06:08:00+00:00" }, { "id": "01a087e3-4d12-73f8-8382-3213956c1f1f", "title": "Use workflows when models cannot follow rules", "url": "https://solmaz.io/x/2086126193582248051/", "published_at": "2026-08-08T16:23:37+00:00" }, { "id": "01a087e3-4d12-73f8-8382-321396320b06", "title": "I'm running some private benchmarks on @liquidai's LFM 2.5 2.6B, and if my results are correct...", "url": "https://solmaz.io/x/2086122623009018302/", "published_at": "2026-08-08T16:09:26+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139641c6fb", "title": "Anyone else use Alibaba’s AACR bench for code review evaluation? github.com/alibaba/aacr-bench", "url": "https://solmaz.io/x/2086049701623906786/", "published_at": "2026-08-08T11:19:40+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139710ac83", "title": "Be not afraid of machines, but of humans wielding the machines", "url": "https://solmaz.io/x/2085966261301989430/", "published_at": "2026-08-08T05:48:06+00:00" }, { "id": "01a087e3-4d12-73f8-8382-3213973118fc", "title": "Obligatory repost of “Clippy” by Gwern", "url": "https://solmaz.io/x/2085778318943756306/", "published_at": "2026-08-07T17:21:17+00:00" }, { "id": "01a087e3-4d12-73f8-8382-321397581222", "title": "Super curious what second hand price gb10's will converge on", "url": "https://solmaz.io/x/2085652776970068084/", "published_at": "2026-08-07T09:02:26+00:00" }, { "id": "01a087e3-4d12-73f8-8382-321397b78957", "title": "Super excited for local.ai!", "url": "https://solmaz.io/x/2085646174984569329/", "published_at": "2026-08-07T08:36:12+00:00" }, { "id": "01a087e3-4d12-73f8-8382-3213988ba756", "title": "OpenClaw should be better in stopping doom loops with smol local models in the next release", "url": "https://solmaz.io/x/2085558448448770097/", "published_at": "2026-08-07T02:47:36+00:00" }, { "id": "01a087e3-4d12-73f8-8382-3213992e81a3", "title": "millions, resetless", "url": "https://solmaz.io/x/2085375242026045492/", "published_at": "2026-08-06T14:39:36+00:00" }, { "id": "01a087e3-4d12-73f8-8382-321399e91f52", "title": "Semantic vendoring keeps customized tools up to date", "url": "https://solmaz.io/x/2085312986881921304/", "published_at": "2026-08-06T10:32:13+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139ac07249", "title": "Pi quality of life extension", "url": "https://solmaz.io/x/2085249284036063488/", "published_at": "2026-08-06T06:19:06+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139ad02680", "title": "Kimi K3 is a valid replacement for GPT 5.6 Sol, subjectively", "url": "https://solmaz.io/x/2085244675297030607/", "published_at": "2026-08-06T06:00:47+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139b13e35a", "title": "My feed is full of people pleading for a codex reset", "url": "https://solmaz.io/x/2085046696338481257/", "published_at": "2026-08-05T16:54:05+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139b6dfe25", "title": "\"What it says, in one breath:\"", "url": "https://solmaz.io/x/2085043630751027420/", "published_at": "2026-08-05T16:41:54+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139bc9bd18", "title": "Model welfare??? I disagree.", "url": "https://solmaz.io/x/2085043297081561408/", "published_at": "2026-08-05T16:40:34+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139c2d16f6", "title": "When model emotions imply model welfare", "url": "https://solmaz.io/x/2084996700158980150/", "published_at": "2026-08-05T13:35:25+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139cadd44e", "title": "they should have gone for overmind, smh", "url": "https://solmaz.io/x/2084850859180274035/", "published_at": "2026-08-05T03:55:54+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139ce15413", "title": "tired: cheering for token plan resets", "url": "https://solmaz.io/x/2084839095097004492/", "published_at": "2026-08-05T03:09:09+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139d7d4f4b", "title": "DeepSeek v4 Flash 0731 surpassed Kimi K3 on Almanbench and scores almost as high as GPT-5.6 Sol...", "url": "https://solmaz.io/x/2084827025685307681/", "published_at": "2026-08-05T02:21:11+00:00" }, { "id": "01a087e3-4d12-73f8-8382-32139e566705", "title": "There will be signs…", "url": "https://solmaz.io/x/2084822625487135087/", "published_at": "2026-08-05T02:03:42+00:00" } ] posts Claim your blog

Back to solmaz.io

Blog · corpus.blog/blogs/solmaz.io/posts

solmaz.io

solmaz.io

2026

unyolo.io fixes this original ↗

13 Aug 2026

😎original ↗

13 Aug 2026

A similar thing can be done with codex session compaction summaries as well btw original ↗

12 Aug 2026

This type of estimations are useful not only for local but for all inference providers original ↗

12 Aug 2026

mfw trying not to think about the 100 models that dropped this week original ↗

11 Aug 2026

Deterministic monitors make long-running agents reliable original ↗

10 Aug 2026

Apparently this went to HN front page briefly last night, so sharing it here as well original ↗

10 Aug 2026

Local AI will favor LPDDR and MoE original ↗

10 Aug 2026

urgh typo, I meant 35-40 tok/s original ↗

9 Aug 2026

Apparently Alibaba did all this work, published a paper, but did not create a public backup of...original ↗

9 Aug 2026

I have made an update to my theoretical upper bound calculation to also predict prefill speed original ↗

9 Aug 2026

Theoretical limits can expose inference problems original ↗

9 Aug 2026

Use workflows when models cannot follow rules original ↗

8 Aug 2026

I'm running some private benchmarks on @liquidai's LFM 2.5 2.6B, and if my results are correct...original ↗

8 Aug 2026

Anyone else use Alibaba’s AACR bench for code review evaluation? github.com/alibaba/aacr-bench original ↗

8 Aug 2026

Be not afraid of machines, but of humans wielding the machines original ↗

8 Aug 2026

Obligatory repost of “Clippy” by Gwern original ↗

7 Aug 2026

Super curious what second hand price gb10's will converge on original ↗

7 Aug 2026

Super excited for local.ai!original ↗

7 Aug 2026

OpenClaw should be better in stopping doom loops with smol local models in the next release original ↗

7 Aug 2026

millions, resetless original ↗

6 Aug 2026

Semantic vendoring keeps customized tools up to date original ↗

6 Aug 2026

Pi quality of life extension original ↗

6 Aug 2026

Kimi K3 is a valid replacement for GPT 5.6 Sol, subjectively original ↗

6 Aug 2026

My feed is full of people pleading for a codex reset original ↗

5 Aug 2026

"What it says, in one breath:"original ↗

5 Aug 2026

Model welfare??? I disagree.original ↗

5 Aug 2026

When model emotions imply model welfare original ↗

5 Aug 2026

they should have gone for overmind, smh original ↗

5 Aug 2026

tired: cheering for token plan resets original ↗

5 Aug 2026

DeepSeek v4 Flash 0731 surpassed Kimi K3 on Almanbench and scores almost as high as GPT-5.6 Sol...original ↗

5 Aug 2026

There will be signs…original ↗

5 Aug 2026